Training confounder-free deep learning models for medical applications

Deep Learning-Guided Dosimetry for Mitigating Local Failure of Patients With Non-Small Cell Lung Cancer Receiving Stereotactic Body Radiation Therapy

PURPOSE

Non-small cell lung cancer (NSCLC) stereotactic body radiation therapy with 50 Gy/5 fractions is sometimes considered controversial, as the nominal biologically effective dose (BED) of 100 Gy is felt by some to be insufficient for long-term local control of some lesions. In this study, we analyzed such patients using explainable deep learning techniques and consequently proposed appropriate treatment planning criteria. These novel criteria could help planners achieve optimized treatment plans for maximal local control.

METHODS AND MATERIALS

A total of 535 patients treated with 50 Gy/5 fractions were used to develop a novel deep learning local response model. A multimodality approach, incorporating computed tomography images, 3-dimensional dose distribution, and patient demographics, combined with a discrete-time survival model, was applied to predict time to failure and the probability of local control. Subsequently, an integrated gradient-weighted class activation mapping method was used to identify the most significant dose-volume metrics predictive of local failure and their optimal cut-points.

RESULTS

The model was cross-validated, showing an acceptable performance (c-index: 0.72, 95% CI, 0.68-0.75); the testing c-index was 0.69. The model's spatial attention was concentrated mostly in the tumors' periphery (planning target volume [PTV] - internal gross target volume [IGTV]) region. Statistically significant dose-volume metrics in improved local control were BED Dnear-min ≥ 103.8 Gy in IGTV (hazard ratio [HR], 0.31; 95% CI, 015-0.63), V104 ≥ 98% in IGTV (HR, 0.30; 95% CI, 0.15-0.60), gEUD ≥ 103.8 Gy in PTV-IGTV (HR, 0.25; 95% CI, 0.12-0.50), and Dmean ≥ 104.5 Gy in PTV-IGTV (HR, 0.25; 95% CI, 0.12-0.51).

CONCLUSIONS

Deep learning-identified dose-volume metrics have shown significant prognostic power (log-rank, P = .003) and could be used as additional actionable criteria for treatment planning in NSCLC stereotactic body radiation therapy patients receiving 50 Gy in 5 fractions. Although our data do not confirm or refute that a significantly higher BED for the prescription dose is necessary for tumor control in NSCLC, it might be clinically effective to escalate the nominal prescribed dose from BED 100 to 105 Gy.

Progress and trends in neurological disorders research based on deep learning

In recent years, deep learning (DL) has emerged as a powerful tool in clinical imaging, offering unprecedented opportunities for the diagnosis and treatment of neurological disorders (NDs). This comprehensive review explores the multifaceted role of DL techniques in leveraging vast datasets to advance our understanding of NDs and improve clinical outcomes. Beginning with a systematic literature review, we delve into the utilization of DL, particularly focusing on multimodal neuroimaging data analysis-a domain that has witnessed rapid progress and garnered significant scientific interest. Our study categorizes and critically analyses numerous DL models, including Convolutional Neural Networks (CNNs), LSTM-CNN, GAN, and VGG, to understand their performance across different types of Neurology Diseases. Through particular analysis, we identify key benchmarks and datasets utilized in training and testing DL models, shedding light on the challenges and opportunities in clinical neuroimaging research. Moreover, we discuss the effectiveness of DL in real-world clinical scenarios, emphasizing its potential to revolutionize ND diagnosis and therapy. By synthesizing existing literature and describing future directions, this review not only provides insights into the current state of DL applications in ND analysis but also covers the way for the development of more efficient and accessible DL techniques. Finally, our findings underscore the transformative impact of DL in reshaping the landscape of clinical neuroimaging, offering hope for enhanced patient care and groundbreaking discoveries in the field of neurology. This review paper is beneficial for neuropathologists and new researchers in this field.

Brain Age Analysis and Dementia Classification using Convolutional Neural Networks trained on Diffusion MRI: Tests in Indian and North American Cohorts

Deep learning models based on convolutional neural networks (CNNs) have been used to classify Alzheimer's disease or infer dementia severity from T1-weighted brain MRI scans. Here, we examine the value of adding diffusion-weighted MRI (dMRI) as an input to these models. Much research in this area focuses on specific datasets such as the Alzheimer's Disease Neuroimaging Initiative (ADNI), which assesses people of North American, largely European ancestry, so we examine how models trained on ADNI, generalize to a new population dataset from India (the NIMHANS cohort). We first benchmark our models by predicting 'brain age' - the task of predicting a person's chronological age from their MRI scan and proceed to AD classification. We also evaluate the benefit of using a 3D CycleGAN approach to harmonize the imaging datasets before training the CNN models. Our experiments show that classification performance improves after harmonization in most cases, as well as better performance for dMRI as input.

Pseudo-class part prototype networks for interpretable breast cancer classification

Interpretability in machine learning has become increasingly important as machine learning is being used in more and more applications, including those with high-stakes consequences such as healthcare where Interpretability has been regarded as a key to the successful adoption of machine learning models. However, using confounding/irrelevant information in making predictions by deep learning models, even the interpretable ones, poses critical challenges to their clinical acceptance. That has recently drawn researchers' attention to issues beyond the mere interpretation of deep learning models. In this paper, we first investigate application of an inherently interpretable prototype-based architecture, known as ProtoPNet, for breast cancer classification in digital pathology and highlight its shortcomings in this application. Then, we propose a new method that uses more medically relevant information and makes more accurate and interpretable predictions. Our method leverages the clustering concept and implicitly increases the number of classes in the training dataset. The proposed method learns more relevant prototypes without any pixel-level annotated data. To have a more holistic assessment, in addition to classification accuracy, we define a new metric for assessing the degree of interpretability based on the comments of a group of skilled pathologists. Experimental results on the BreakHis dataset show that the proposed method effectively improves the classification accuracy and interpretability by respectively 8 % and 18 % . Therefore, the proposed method can be seen as a step toward implementing interpretable deep learning models for the detection of breast cancer using histopathology images.

Increases in regional brain volume across two native South American male populations

Industrialized environments, despite benefits such as higher levels of formal education and lower rates of infections, can also have pernicious impacts upon brain atrophy. Partly for this reason, comparing age-related brain volume trajectories between industrialized and non-industrialized populations can help to suggest lifestyle correlates of brain health. The Tsimane, indigenous to the Bolivian Amazon, derive their subsistence from foraging and horticulture and are physically active. The Moseten, a mixed-ethnicity farming population, are physically active but less than the Tsimane. Within both populations (N = 1024; age range = 46-83), we calculated regional brain volumes from computed tomography and compared their cross-sectional trends with age to those of UK Biobank (UKBB) participants (N = 19,973; same age range). Surprisingly among Tsimane and Moseten (T/M) males, some parietal and occipital structures mediating visuospatial abilities exhibit small but significant increases in regional volume with age. UKBB males exhibit a steeper negative trend of regional volume with age in frontal and temporal structures compared to T/M males. However, T/M females exhibit significantly steeper rates of brain volume decrease with age compared to UKBB females, particularly for some cerebro-cortical structures (e.g., left subparietal cortex). Across the three populations, observed trends exhibit no interhemispheric asymmetry. In conclusion, the age-related rate of regional brain volume change may differ by lifestyle and sex. The lack of brain volume reduction with age is not known to exist in other human population, highlighting the putative role of lifestyle in constraining regional brain atrophy and promoting elements of non-industrialized lifestyle like higher physical activity.

Predicting clinical outcomes of SARS-CoV-2 infection during the Omicron wave using machine learning

The Omicron SARS-CoV-2 variant continues to strain healthcare systems. Developing tools that facilitate the identification of patients at highest risk of adverse outcomes is a priority. The study objectives are to develop population-scale predictive models that: 1) identify predictors of adverse outcomes with Omicron surge SARS-CoV-2 infections, and 2) predict the impact of prioritized vaccination of high-risk groups for said outcome. We prepared a retrospective longitudinal observational study of a national cohort of 172,814 patients in the U.S. Veteran Health Administration who tested positive for SARS-CoV-2 from January 15 to August 15, 2022. We utilized sociodemographic characteristics, comorbidities, and vaccination status, at time of testing positive for SARS-CoV-2 to predict hospitalization, escalation of care (high-flow oxygen, mechanical ventilation, vasopressor use, dialysis, or extracorporeal membrane oxygenation), and death within 30 days. Machine learning models demonstrated that advanced age, high comorbidity burden, lower body mass index, unvaccinated status, and oral anticoagulant use were the important predictors of hospitalization and escalation of care. Similar factors predicted death. However, anticoagulant use did not predict mortality risk. The all-cause death model showed the highest discrimination (Area Under the Curve (AUC) = 0.903, 95% Confidence Interval (CI): 0.895, 0.911) followed by hospitalization (AUC = 0.822, CI: 0.818, 0.826), then escalation of care (AUC = 0.793, CI: 0.784, 0.805). Assuming a vaccine efficacy range of 70.8 to 78.7%, our simulations projected that targeted prevention in the highest risk group may have reduced 30-day hospitalization and death in more than 2 of 5 unvaccinated patients.

The path from task-specific to general purpose artificial intelligence for medical diagnostics: A bibliometric analysis

Artificial intelligence (AI) has revolutionized many fields, and its potential in healthcare has been increasingly recognized. Based on diverse data sources such as imaging, laboratory tests, medical records, and electrophysiological data, diagnostic AI has witnessed rapid development in recent years. A comprehensive understanding of the development status, contributing factors, and their relationships in the application of AI to medical diagnostics is essential to further promote its use in clinical practice. In this study, we conducted a bibliometric analysis to explore the evolution of task-specific to general-purpose AI for medical diagnostics. We used the Web of Science database to search for relevant articles published between 2010 and 2023, and applied VOSviewer, the R package Bibliometrix, and CiteSpace to analyze collaborative networks and keywords. Our analysis revealed that the field of AI in medical diagnostics has experienced rapid growth in recent years, with a focus on tasks such as image analysis, disease prediction, and decision support. Collaborative networks were observed among researchers and institutions, indicating a trend of global cooperation in this field. Additionally, we identified several key factors contributing to the development of AI in medical diagnostics, including data quality, algorithm design, and computational power. Challenges to progress in the field include model explainability, robustness, and equality, which will require multi-stakeholder, interdisciplinary collaboration to tackle. Our study provides a holistic understanding of the path from task-specific, mono-modal AI toward general-purpose, multimodal AI for medical diagnostics. With the continuous improvement of AI technology and the accumulation of medical data, we believe that AI will play a greater role in medical diagnostics in the future.

Transparent medical image AI via an image-text foundation model grounded in medical literature

Building trustworthy and transparent image-based medical artificial intelligence (AI) systems requires the ability to interrogate data and models at all stages of the development pipeline, from training models to post-deployment monitoring. Ideally, the data and associated AI systems could be described using terms already familiar to physicians, but this requires medical datasets densely annotated with semantically meaningful concepts. In the present study, we present a foundation model approach, named MONET (medical concept retriever), which learns how to connect medical images with text and densely scores images on concept presence to enable important tasks in medical AI development and deployment such as data auditing, model auditing and model interpretation. Dermatology provides a demanding use case for the versatility of MONET, due to the heterogeneity in diseases, skin tones and imaging modalities. We trained MONET based on 105,550 dermatological images paired with natural language descriptions from a large collection of medical literature. MONET can accurately annotate concepts across dermatology images as verified by board-certified dermatologists, competitively with supervised models built on previously concept-annotated dermatology datasets of clinical images. We demonstrate how MONET enables AI transparency across the entire AI system development pipeline, from building inherently interpretable models to dataset and model auditing, including a case study dissecting the results of an AI clinical trial.

Residual Partial Least Squares Learning: Brain Cortical Thickness Simultaneously Predicts Eight Non-pairwise-correlated Behavioural and Disease Outcomes in Alzheimer's Disease

Alzheimer's Disease (AD) is the leading cause of dementia. It results in cortical thickness changes and is associated with a decline in cognition and behaviour. Such decline affects multiple important day-to-day functions, including memory, language, orientation, judgment and problem-solving. Recent research has made important progress in identifying brain regions associated with single outcomes, such as individual AD status and general cognitive decline. The complex projection from multiple brain areas to multiple AD outcomes, however, remains poorly understood. This makes the assessment and especially the prediction of multiple AD outcomes - each of which may unveil an integral yet different aspect of the disease - challenging, particularly when some are not strongly correlated. Here, uniting residual learning, partial least squares (PLS), and predictive modelling, we develop an explainable, generalisable, and reproducible method called the Residual Partial Least Squares Learning (the re-PLS Learning) to (1) chart the pathways between large-scale multivariate brain cortical thickness data (inputs) and multivariate disease and behaviour data (outcomes); (2) simultaneously predict multiple, non-pairwise-correlated outcomes; (3) control for confounding variables (e.g., age and gender) affecting both inputs and outcomes and the pathways in-between; (4) perform longitudinal AD disease status classification and disease severity prediction. We evaluate the performance of the proposed method against a variety of alternatives on data from AD patients, subjects with mild cognitive impairment (MCI), and cognitively normal individuals ( n = 1,196 ) from the Alzheimer's Disease Neuroimaging Initiative (ADNI). Our results unveil pockets of brain areas in the temporal, frontal, sensorimotor, and cingulate areas whose cortical thickness may be respectively associated with declines in different cognitive and behavioural subdomains in AD. Finally, we characterise re-PLS' geometric interpretation and mathematical support for delivering meaningful neurobiological insights and provide an open software package (re-PLS) available at https://github.com/thanhvd18/rePLS.

Detection of COVID-19 features in lung ultrasound images using deep neural networks

BACKGROUND

Deep neural networks (DNNs) to detect COVID-19 features in lung ultrasound B-mode images have primarily relied on either in vivo or simulated images as training data. However, in vivo images suffer from limited access to required manual labeling of thousands of training image examples, and simulated images can suffer from poor generalizability to in vivo images due to domain differences. We address these limitations and identify the best training strategy.

METHODS

We investigated in vivo COVID-19 feature detection with DNNs trained on our carefully simulated datasets (40,000 images), publicly available in vivo datasets (174 images), in vivo datasets curated by our team (958 images), and a combination of simulated and internal or external in vivo datasets. Seven DNN training strategies were tested on in vivo B-mode images from COVID-19 patients.

RESULTS

Here, we show that Dice similarity coefficients (DSCs) between ground truth and DNN predictions are maximized when simulated data are mixed with external in vivo data and tested on internal in vivo data (i.e., 0.482 ± 0.211), compared with using only simulated B-mode image training data (i.e., 0.464 ± 0.230) or only external in vivo B-mode training data (i.e., 0.407 ± 0.177). Additional maximization is achieved when a separate subset of the internal in vivo B-mode images are included in the training dataset, with the greatest maximization of DSC (and minimization of required training time, or epochs) obtained after mixing simulated data with internal and external in vivo data during training, then testing on the held-out subset of the internal in vivo dataset (i.e., 0.735 ± 0.187).

CONCLUSIONS

DNNs trained with simulated and in vivo data are promising alternatives to training with only real or only simulated data when segmenting in vivo COVID-19 lung ultrasound features.

Non-neotissue constituents as underestimated confounders in the assessment of tissue engineered constructs by near-infrared spectroscopy

Non-destructive assessments are required for the quality control of tissue-engineered constructs and the optimization of the tissue culture process. Near-infrared (NIR) spectroscopy coupled with machine learning (ML) provides a promising approach for such assessment. However, due to its nonspecific nature, each spectrum incorporates information on both neotissue and non-neotissue constituents of the construct; the effect of these constituents on the NIR-based assessments of tissue-engineered constructs has been overlooked in previous studies. This study investigates the effect of scaffolds, growth factors, and buffers on NIR-based assessments of tissue-engineered constructs. To determine if these non-neotissue constituents have a measurable effect on the NIR spectra of the constructs that can introduce bias in their assessment, nine ML algorithms were evaluated in classifying the NIR spectra of engineered cartilage according to the scaffold used to prepare the constructs, the growth factors added to the culture media, and the buffers used for storing the constructs. The effect of controlling for these constituents was also evaluated using controlled and uncontrolled NIR-based ML models for predicting tissue maturity as an example of neotissue-related properties of interest. Samples used in this study were prepared using norbornene-modified hyaluronic acid scaffolds with or without the conjugation of an N-cadherin mimetic peptide. Selected samples were supplemented with transforming growth factor-beta1 or bone morphogenetic protein-9 growth factor. Some samples were frozen in cell lysis buffer, while the remaining samples were frozen in PBS until required for NIR analysis. The ML models for classifying the spectra of the constructs according to the four constituents exhibited high to fair performances, with F1 scores ranging from 0.9 to 0.52. Moreover, controlling for the four constituents significantly improved the performance of the models for predicting tissue maturity, with improvement in F1 scores ranging from 0.09 to 0.77. In conclusion, non-neotissue constituents have measurable effects on the NIR spectra of tissue-engineered constructs that can be detected by ML algorithms and introduce bias in the assessment of the constructs by NIR spectroscopy. Therefore, controlling for these constituents is necessary for reliable NIR-based assessments of tissue-engineered constructs.

A primer on the use of machine learning to distil knowledge from data in biological psychiatry

Applications of machine learning in the biomedical sciences are growing rapidly. This growth has been spurred by diverse cross-institutional and interdisciplinary collaborations, public availability of large datasets, an increase in the accessibility of analytic routines, and the availability of powerful computing resources. With this increased access and exposure to machine learning comes a responsibility for education and a deeper understanding of its bases and bounds, borne equally by data scientists seeking to ply their analytic wares in medical research and by biomedical scientists seeking to harness such methods to glean knowledge from data. This article provides an accessible and critical review of machine learning for a biomedically informed audience, as well as its applications in psychiatry. The review covers definitions and expositions of commonly used machine learning methods, and historical trends of their use in psychiatry. We also provide a set of standards, namely Guidelines for REporting Machine Learning Investigations in Neuropsychiatry (GREMLIN), for designing and reporting studies that use machine learning as a primary data-analysis approach. Lastly, we propose the establishment of the Machine Learning in Psychiatry (MLPsych) Consortium, enumerate its objectives, and identify areas of opportunity for future applications of machine learning in biological psychiatry. This review serves as a cautiously optimistic primer on machine learning for those on the precipice as they prepare to dive into the field, either as methodological practitioners or well-informed consumers.

AI-based X-ray fracture analysis of the distal radius: accuracy between representative classification, detection and segmentation deep learning models for clinical practice

OBJECTIVES

To aid in selecting the optimal artificial intelligence (AI) solution for clinical application, we directly compared performances of selected representative custom-trained or commercial classification, detection and segmentation models for fracture detection on musculoskeletal radiographs of the distal radius by aligning their outputs.

DESIGN AND SETTING

This single-centre retrospective study was conducted on a random subset of emergency department radiographs from 2008 to 2018 of the distal radius in Germany.

MATERIALS AND METHODS

An image set was created to be compatible with training and testing classification and segmentation models by annotating examinations for fractures and overlaying fracture masks, if applicable. Representative classification and segmentation models were trained on 80% of the data. After output binarisation, their derived fracture detection performances as well as that of a standard commercially available solution were compared on the remaining X-rays (20%) using mainly accuracy and area under the receiver operating characteristic (AUROC).

RESULTS

A total of 2856 examinations with 712 (24.9%) fractures were included in the analysis. Accuracies reached up to 0.97 for the classification model, 0.94 for the segmentation model and 0.95 for BoneView. Cohen's kappa was at least 0.80 in pairwise comparisons, while Fleiss' kappa was 0.83 for all models. Fracture predictions were visualised with all three methods at different levels of detail, ranking from downsampled image region for classification over bounding box for detection to single pixel-level delineation for segmentation.

CONCLUSIONS

All three investigated approaches reached high performances for detection of distal radius fractures with simple preprocessing and postprocessing protocols on the custom-trained models. Despite their underlying structural differences, selection of one's fracture analysis AI tool in the frame of this study reduces to the desired flavour of automation: automated classification, AI-assisted manual fracture reading or minimised false negatives.

An AI-Enabled Bias-Free Respiratory Disease Diagnosis Model Using Cough Audio

Cough-based diagnosis for respiratory diseases (RDs) using artificial intelligence (AI) has attracted considerable attention, yet many existing studies overlook confounding variables in their predictive models. These variables can distort the relationship between cough recordings (input data) and RD status (output variable), leading to biased associations and unrealistic model performance. To address this gap, we propose the Bias-Free Network (RBF-Net), an end-to-end solution that effectively mitigates the impact of confounders in the training data distribution. RBF-Net ensures accurate and unbiased RD diagnosis features, emphasizing its relevance by incorporating a COVID-19 dataset in this study. This approach aims to enhance the reliability of AI-based RD diagnosis models by navigating the challenges posed by confounding variables. A hybrid of a Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM) networks is proposed for the feature encoder module of RBF-Net. An additional bias predictor is incorporated in the classification scheme to formulate a conditional Generative Adversarial Network (c-GAN) that helps in decorrelating the impact of confounding variables from RD prediction. The merit of RBF-Net is demonstrated by comparing classification performance with a State-of-The-Art (SoTA) Deep Learning (DL) model (CNN-LSTM) after training on different unbalanced COVID-19 data sets, created by using a large-scale proprietary cough data set. RBF-Net proved its robustness against extremely biased training scenarios by achieving test set accuracies of 84.1%, 84.6%, and 80.5% for the following confounding variables-gender, age, and smoking status, respectively. RBF-Net outperforms the CNN-LSTM model test set accuracies by 5.5%, 7.7%, and 8.2%, respectively.

Assessment of inspiration and technical quality in anteroposterior thoracic radiographs using machine learning

INTRODUCTION

Chest radiographs are the most performed radiographic procedure, but suboptimal technical factors can impact clinical interpretation. A deep learning model was developed to assess technical and inspiratory adequacy of anteroposterior chest radiographs.

METHODS

Adult anteroposterior chest radiographs (n = 2375) were assessed for technical adequacy, and if otherwise technically adequate, for adequacy of inspiration. Images were labelled by an experienced radiologist with one of three ground truth labels: inadequate technique (n = 605, 25.5 %), adequate inspiration (n = 900, 37.9 %), and inadequate inspiration (n = 870, 36.6 %). A convolutional neural network was then iteratively trained to predict these labels and evaluated using recall, precision, F1 and micro-F1, and Gradient-weighted Class Activation Mapping analysis on a hold-out test set. Impact of kyphosis on model accuracy was assessed.

RESULTS

The model performed best for radiographs with adequate technique, and worst for images with inadequate technique. Recall was highest (89 %) for radiographs with both adequate technique and inspiration, with recall of 81 % for images with adequate technique and inadequate inspiration, and 60 % for images with inadequate technique, although precision was highest (85 %) for this category. Per-class F1 was 80 %, 81 % and 70 % for adequate inspiration, inadequate inspiration, and inadequate technique respectively. Weighted F1 and Micro F1 scores were 78 %. Presence or absence of kyphosis had no significant impact on model accuracy in images with adequate technique.

CONCLUSION

This study explores the promising performance of a machine learning algorithm for assessment of inspiratory adequacy and overall technical adequacy for anteroposterior chest radiograph acquisition.

IMPLICATIONS FOR PRACTICE

With further refinement, machine learning can contribute to education and quality improvement in radiology departments.

Improving model fairness in image-based computer-aided diagnosis

Deep learning has become a popular tool for computer-aided diagnosis using medical images, sometimes matching or exceeding the performance of clinicians. However, these models can also reflect and amplify human bias, potentially resulting inaccurate missed diagnoses. Despite this concern, the problem of improving model fairness in medical image classification by deep learning has yet to be fully studied. To address this issue, we propose an algorithm that leverages the marginal pairwise equal opportunity to reduce bias in medical image classification. Our evaluations across four tasks using four independent large-scale cohorts demonstrate that our proposed algorithm not only improves fairness in individual and intersectional subgroups but also maintains overall performance. Specifically, the relative change in pairwise fairness difference between our proposed model and the baseline model was reduced by over 35%, while the relative change in AUC value was typically within 1%. By reducing the bias generated by deep learning models, our proposed approach can potentially alleviate concerns about the fairness and reliability of image-based computer-aided diagnosis.

A translational perspective towards clinical AI fairness

Artificial intelligence (AI) has demonstrated the ability to extract insights from data, but the fairness of such data-driven insights remains a concern in high-stakes fields. Despite extensive developments, issues of AI fairness in clinical contexts have not been adequately addressed. A fair model is normally expected to perform equally across subgroups defined by sensitive variables (e.g., age, gender/sex, race/ethnicity, socio-economic status, etc.). Various fairness measurements have been developed to detect differences between subgroups as evidence of bias, and bias mitigation methods are designed to reduce the differences detected. This perspective of fairness, however, is misaligned with some key considerations in clinical contexts. The set of sensitive variables used in healthcare applications must be carefully examined for relevance and justified by clear clinical motivations. In addition, clinical AI fairness should closely investigate the ethical implications of fairness measurements (e.g., potential conflicts between group- and individual-level fairness) to select suitable and objective metrics. Generally defining AI fairness as "equality" is not necessarily reasonable in clinical settings, as differences may have clinical justifications and do not indicate biases. Instead, "equity" would be an appropriate objective of clinical AI fairness. Moreover, clinical feedback is essential to developing fair and well-performing AI models, and efforts should be made to actively involve clinicians in the process. The adaptation of AI fairness towards healthcare is not self-evident due to misalignments between technical developments and clinical considerations. Multidisciplinary collaboration between AI researchers, clinicians, and ethicists is necessary to bridge the gap and translate AI fairness into real-life benefits.

Development of Debiasing Technique for Lung Nodule Chest X-ray Datasets to Generalize Deep Learning Models

Screening programs for early lung cancer diagnosis are uncommon, primarily due to the challenge of reaching at-risk patients located in rural areas far from medical facilities. To overcome this obstacle, a comprehensive approach is needed that combines mobility, low cost, speed, accuracy, and privacy. One potential solution lies in combining the chest X-ray imaging mode with federated deep learning, ensuring that no single data source can bias the model adversely. This study presents a pre-processing pipeline designed to debias chest X-ray images, thereby enhancing internal classification and external generalization. The pipeline employs a pruning mechanism to train a deep learning model for nodule detection, utilizing the most informative images from a publicly available lung nodule X-ray dataset. Histogram equalization is used to remove systematic differences in image brightness and contrast. Model training is then performed using combinations of lung field segmentation, close cropping, and rib/bone suppression. The resulting deep learning models, generated through this pre-processing pipeline, demonstrate successful generalization on an independent lung nodule dataset. By eliminating confounding variables in chest X-ray images and suppressing signal noise from the bone structures, the proposed deep learning lung nodule detection algorithm achieves an external generalization accuracy of 89%. This approach paves the way for the development of a low-cost and accessible deep learning-based clinical system for lung cancer screening.

Penalized decomposition using residuals (PeDecURe) for feature extraction in the presence of nuisance variables

Neuroimaging data are an increasingly important part of etiological studies of neurological and psychiatric disorders. However, mitigating the influence of nuisance variables, including confounders, remains a challenge in image analysis. In studies of Alzheimer's disease, for example, an imbalance in disease rates by age and sex may make it difficult to distinguish between structural patterns in the brain (as measured by neuroimaging scans) attributable to disease progression and those characteristic of typical human aging or sex differences. Concerningly, when not properly accounted for, nuisance variables pose threats to the generalizability and interpretability of findings from these studies. Motivated by this critical issue, in this work, we examine the impact of nuisance variables on feature extraction methods and propose Penalized Decomposition Using Residuals (PeDecURe), a new method for obtaining nuisance variable-adjusted features. PeDecURe estimates primary directions of variation which maximize covariance between partially residualized imaging features and a variable of interest (e.g., Alzheimer's diagnosis) while simultaneously mitigating the influence of nuisance variation through a penalty on the covariance between partially residualized imaging features and those variables. Using features derived using PeDecURe's first direction of variation, we train a highly accurate and generalizable predictive model, as evidenced by its robustness in testing samples with different underlying nuisance variable distributions. We compare PeDecURe to commonly used decomposition methods (principal component analysis (PCA) and partial least squares) as well as a confounder-adjusted variation of PCA. We find that features derived from PeDecURe offer greater accuracy and generalizability and lower correlations with nuisance variables compared with the other methods. While PeDecURe is primarily motivated by challenges that arise in the analysis of neuroimaging data, it is broadly applicable to data sets with highly correlated features, where novel methods to handle nuisance variables are warranted.

Challenges of implementing computer-aided diagnostic models for neuroimages in a clinical setting

Advances in artificial intelligence have cultivated a strong interest in developing and validating the clinical utilities of computer-aided diagnostic models. Machine learning for diagnostic neuroimaging has often been applied to detect psychological and neurological disorders, typically on small-scale datasets or data collected in a research setting. With the collection and collation of an ever-growing number of public datasets that researchers can freely access, much work has been done in adapting machine learning models to classify these neuroimages by diseases such as Alzheimer's, ADHD, autism, bipolar disorder, and so on. These studies often come with the promise of being implemented clinically, but despite intense interest in this topic in the laboratory, limited progress has been made in clinical implementation. In this review, we analyze challenges specific to the clinical implementation of diagnostic AI models for neuroimaging data, looking at the differences between laboratory and clinical settings, the inherent limitations of diagnostic AI, and the different incentives and skill sets between research institutions, technology companies, and hospitals. These complexities need to be recognized in the translation of diagnostic AI for neuroimaging from the laboratory to the clinic.

Individualized Risk Prediction for Improved Chronic Wound Management

Significance: Chronic wounds are associated with significant morbidity, marked loss of quality of life, and considerable economic burden. Evidence-based risk prediction to guide improved wound prevention and treatment is limited by the complexity in their etiology, clinical underreporting, and a lack of studies using large high-quality datasets. Recent Advancements: The objective of this review is to summarize key components and challenges in the development of personalized risk prediction tools for both prevention and management of chronic wounds, while highlighting several innovations in the development of better risk stratification. Critical Issues: Regression-based risk prediction approaches remain important for assessment of prognosis and risk stratification in chronic wound management. Advances in statistical computing have boosted the development of several promising machine learning (ML) and other semiautomated classification tools. These methods may be better placed to handle large number of wound healing risk factors from large datasets, potentially resulting in better risk prediction when combined with conventional methods and clinical experience and expertise. Future Directions: Where the number of predictors is large and heterogenous, the correlations between various risk factors complex, and very large data sets are available, ML may prove a powerful adjuvant for risk stratifying patients predisposed to chronic wounds. Conventional regression-based approaches remain important, particularly where the number of predictors is relatively small. Translating estimated risk derived from ML algorithms into practical prediction tools for use in clinical practice remains challenging.

Fostering transparent medical image AI via an image-text foundation model grounded in medical literature

Building trustworthy and transparent image-based medical AI systems requires the ability to interrogate data and models at all stages of the development pipeline: from training models to post-deployment monitoring. Ideally, the data and associated AI systems could be described using terms already familiar to physicians, but this requires medical datasets densely annotated with semantically meaningful concepts. Here, we present a foundation model approach, named MONET (Medical cONcept rETriever), which learns how to connect medical images with text and generates dense concept annotations to enable tasks in AI transparency from model auditing to model interpretation. Dermatology provides a demanding use case for the versatility of MONET, due to the heterogeneity in diseases, skin tones, and imaging modalities. We trained MONET on the basis of 105,550 dermatological images paired with natural language descriptions from a large collection of medical literature. MONET can accurately annotate concepts across dermatology images as verified by board-certified dermatologists, outperforming supervised models built on previously concept-annotated dermatology datasets. We demonstrate how MONET enables AI transparency across the entire AI development pipeline from dataset auditing to model auditing to building inherently interpretable models.

CheckList for EvaluAtion of Radiomics research (CLEAR): a step-by-step reporting guideline for authors and reviewers endorsed by ESR and EuSoMII

Even though radiomics can hold great potential for supporting clinical decision-making, its current use is mostly limited to academic research, without applications in routine clinical practice. The workflow of radiomics is complex due to several methodological steps and nuances, which often leads to inadequate reporting and evaluation, and poor reproducibility. Available reporting guidelines and checklists for artificial intelligence and predictive modeling include relevant good practices, but they are not tailored to radiomic research. There is a clear need for a complete radiomics checklist for study planning, manuscript writing, and evaluation during the review process to facilitate the repeatability and reproducibility of studies. We here present a documentation standard for radiomic research that can guide authors and reviewers. Our motivation is to improve the quality and reliability and, in turn, the reproducibility of radiomic research. We name the checklist CLEAR (CheckList for EvaluAtion of Radiomics research), to convey the idea of being more transparent. With its 58 items, the CLEAR checklist should be considered a standardization tool providing the minimum requirements for presenting clinical radiomics research. In addition to a dynamic online version of the checklist, a public repository has also been set up to allow the radiomics community to comment on the checklist items and adapt the checklist for future versions. Prepared and revised by an international group of experts using a modified Delphi method, we hope the CLEAR checklist will serve well as a single and complete scientific documentation tool for authors and reviewers to improve the radiomics literature.

Machine Learning Approaches to Understand Cognitive Phenotypes in People With HIV

Cognitive disorders are prevalent in people with HIV (PWH) despite antiretroviral therapy. Given the heterogeneity of cognitive disorders in PWH in the current era and evidence that these disorders have different etiologies and risk factors, scientific rationale is growing for using data-driven models to identify biologically defined subtypes (biotypes) of these disorders. Here, we discuss the state of science using machine learning to understand cognitive phenotypes in PWH and their associated comorbidities, biological mechanisms, and risk factors. We also discuss methods, example applications, challenges, and what will be required from the field to successfully incorporate machine learning in research on cognitive disorders in PWH. These topics were discussed at the National Institute of Mental Health meeting on "Biotypes of CNS Complications in People Living with HIV" held in October 2021. These ongoing research initiatives seek to explain the heterogeneity of cognitive phenotypes in PWH and their associated biological mechanisms to facilitate clinical management and tailored interventions.

Spectral imaging enables contrast agent-free real-time ischemia monitoring in laparoscopic surgery

Laparoscopic surgery has evolved as a key technique for cancer diagnosis and therapy. While characterization of the tissue perfusion is crucial in various procedures, such as partial nephrectomy, doing so by means of visual inspection remains highly challenging. We developed a laparoscopic real-time multispectral imaging system featuring a compact and lightweight multispectral camera and the possibility to complement the conventional surgical view of the patient with functional information at a video rate of 25 Hz. To enable contrast agent-free ischemia monitoring during laparoscopic partial nephrectomy, we phrase the problem of ischemia detection as an out-of-distribution detection problem that does not rely on data from any other patient and uses an ensemble of invertible neural networks at its core. An in-human trial demonstrates the feasibility of our approach and highlights the potential of spectral imaging combined with advanced deep learning-based analysis tools for fast, efficient, reliable, and safe functional laparoscopic imaging.

Adversarial confound regression and uncertainty measurements to classify heterogeneous clinical MRI in Mass General Brigham

In this work, we introduce a novel deep learning architecture, MUCRAN (Multi-Confound Regression Adversarial Network), to train a deep learning model on clinical brain MRI while regressing demographic and technical confounding factors. We trained MUCRAN using 17,076 clinical T1 Axial brain MRIs collected from Massachusetts General Hospital before 2019 and demonstrated that MUCRAN could successfully regress major confounding factors in the vast clinical dataset. We also applied a method for quantifying uncertainty across an ensemble of these models to automatically exclude out-of-distribution data in AD detection. By combining MUCRAN and the uncertainty quantification method, we showed consistent and significant increases in the AD detection accuracy for newly collected MGH data (post-2019; 84.6% with MUCRAN vs. 72.5% without MUCRAN) and for data from other hospitals (90.3% from Brigham and Women's Hospital and 81.0% from other hospitals). MUCRAN offers a generalizable approach for deep-learning-based disease detection in heterogenous clinical data.

Equitable modelling of brain imaging by counterfactual augmentation with morphologically constrained 3D deep generative models

We describe CounterSynth, a conditional generative model of diffeomorphic deformations that induce label-driven, biologically plausible changes in volumetric brain images. The model is intended to synthesise counterfactual training data augmentations for downstream discriminative modelling tasks where fidelity is limited by data imbalance, distributional instability, confounding, or underspecification, and exhibits inequitable performance across distinct subpopulations. Focusing on demographic attributes, we evaluate the quality of synthesised counterfactuals with voxel-based morphometry, classification and regression of the conditioning attributes, and the Fréchet inception distance. Examining downstream discriminative performance in the context of engineered demographic imbalance and confounding, we use UK Biobank and OASIS magnetic resonance imaging data to benchmark CounterSynth augmentation against current solutions to these problems. We achieve state-of-the-art improvements, both in overall fidelity and equity. The source code for CounterSynth is available at https://github.com/guilherme-pombo/CounterSynth.

General psychopathology factor (p-factor) prediction using resting-state functional connectivity and a scanner-generalization neural network

The general psychopathology factor (p-factor) represents shared variance across mental disorders based on psychopathologic symptoms. The Adolescent Brain Cognitive Development (ABCD) Study offers an unprecedented opportunity to investigate functional networks (FNs) from functional magnetic resonance imaging (fMRI) associated with the psychopathology of an adolescent cohort (n > 10,000). However, the heterogeneities associated with the use of multiple sites and multiple scanners in the ABCD Study need to be overcome to improve the prediction of the p-factor using fMRI. We proposed a scanner-generalization neural network (SGNN) to predict the individual p-factor by systematically reducing the scanner effect for resting-state functional connectivity (RSFC). We included 6905 adolescents from 18 sites whose fMRI data were collected using either Siemens or GE scanners. The p-factor was estimated based on the Child Behavior Checklist (CBCL) scores available in the ABCD study using exploratory factor analysis. We evaluated the Pearson's correlation coefficients (CCs) for p-factor prediction via leave-one/two-site-out cross-validation (LOSOCV/LTSOCV) and identified important FNs from the weight features (WFs) of the SGNN. The CCs were higher for the SGNN than for alternative models when using both LOSOCV (0.1631 ± 0.0673 for the SGNN vs. 0.1497 ± 0.0710 for kernel ridge regression [KRR]; p < 0.05 from a two-tailed paired t-test) and LTSOCV (0.1469 ± 0.0381 for the SGNN vs. 0.1394 ± 0.0359 for KRR; p = 0.01). It was found that (a) the default-mode and dorsal attention FNs were important for p-factor prediction, and (b) the intra-visual FN was important for scanner generalization. We demonstrated the efficacy of our novel SGNN model for p-factor prediction while simultaneously eliminating scanner-related confounding effects for RSFC.

Machine learning seizure prediction: one problematic but accepted practice

Objective.Epilepsy is one of the most common neurological disorders and can have a devastating effect on a person's quality of life. As such, the search for markers which indicate an upcoming seizure is a critically important area of research which would allow either on-demand treatment or early warning for people suffering with these disorders. There is a growing body of work which uses machine learning methods to detect pre-seizure biomarkers from electroencephalography (EEG), however the high prediction rates published do not translate into the clinical setting. Our objective is to investigate a potential reason for this.Approach.We conduct an empirical study of a commonly used data labelling method for EEG seizure prediction which relies on labelling small windows of EEG data in temporal groups then selecting randomly from those windows to validate results. We investigate a confound for this approach for seizure prediction and demonstrate the ease at which it can be inadvertently learned by a machine learning system.Main results.We find that non-seizure signals can create decision surfaces for machine learning approaches which can result in false high prediction accuracy on validation datasets. We prove this by training an artificial neural network to learn fake seizures (fully decoupled from biology) in real EEG.Significance.The significance of our findings is that many existing works may be reporting results based on this confound and that future work should adhere to stricter requirements in mitigating this confound. The problematic, but commonly accepted approach in the literature for seizure prediction labelling is potentially preventing real advances in developing solutions for these sufferers. By adhering to the guidelines in this paper future work in machine learning seizure prediction is more likely to be clinically relevant.

Untargeted Metabolomics for Disease-Specific Signatures

Human diseases account for complex traits that usually exhibit markedly diverse clinical manifestations coming from a series of pathogenic processes that shape heterogeneous phenotypes. Considering that correlation does not imply causation as well as population differences and/or inter-individual variability, disease-specific signatures are becoming critical for biomarker discovery. Untargeted metabolomics is deemed to be a powerful approach to delineate molecular pathways of prime interest. Metabotypes capture the interplay of genomics and environmental influences per se. Untargeted metabolomics share the charm of being not only hypothesis-driven but also hypothesis-generating. Notwithstanding, the applicability of untargeted metabolomics toward clinically relevant outcomes depend on wet- and dry-lab procedures in the context of elegant study designs with clear rationale. As ideal may be far from feasible, herein we provide recommendations to combat sample mishandling that adversely affect data outcomes and if so, deal with imbalanced datasets toward data integrity.

Confound-leakage: confound removal in machine learning leads to leakage

BACKGROUND

Machine learning (ML) approaches are a crucial component of modern data analysis in many fields, including epidemiology and medicine. Nonlinear ML methods often achieve accurate predictions, for instance, in personalized medicine, as they are capable of modeling complex relationships between features and the target. Problematically, ML models and their predictions can be biased by confounding information present in the features. To remove this spurious signal, researchers often employ featurewise linear confound regression (CR). While this is considered a standard approach for dealing with confounding, possible pitfalls of using CR in ML pipelines are not fully understood.

RESULTS

We provide new evidence that, contrary to general expectations, linear confound regression can increase the risk of confounding when combined with nonlinear ML approaches. Using a simple framework that uses the target as a confound, we show that information leaked via CR can increase null or moderate effects to near-perfect prediction. By shuffling the features, we provide evidence that this increase is indeed due to confound-leakage and not due to revealing of information. We then demonstrate the danger of confound-leakage in a real-world clinical application where the accuracy of predicting attention-deficit/hyperactivity disorder is overestimated using speech-derived features when using depression as a confound.

CONCLUSIONS

Mishandling or even amplifying confounding effects when building ML models due to confound-leakage, as shown, can lead to untrustworthy, biased, and unfair predictions. Our expose of the confound-leakage pitfall and provided guidelines for dealing with it can help create more robust and trustworthy ML models.

Confounds in neuroimaging: A clear case of sex as a confound in brain-based prediction

Muscle weakness is common in many neurological, neuromuscular, and musculoskeletal conditions. Muscle size only partially explains muscle strength as adaptions within the nervous system also contribute to strength. Brain-based biomarkers of neuromuscular function could provide diagnostic, prognostic, and predictive value in treating these disorders. Therefore, we sought to characterize and quantify the brain's contribution to strength by developing multimodal MRI pipelines to predict grip strength. However, the prediction of strength was not straightforward, and we present a case of sex being a clear confound in brain decoding analyses. While each MRI modality-structural MRI (i.e., gray matter morphometry), diffusion MRI (i.e., white matter fractional anisotropy), resting state functional MRI (i.e., functional connectivity), and task-evoked functional MRI (i.e., left or right hand motor task activation)-and a multimodal prediction pipeline demonstrated significant predictive power for strength (R ² = 0.108-0.536, p ≤ 0.001), after correcting for sex, the predictive power was substantially reduced (R ² = -0.038-0.075). Next, we flipped the analysis and demonstrated that each MRI modality and a multimodal prediction pipeline could significantly predict sex (accuracy = 68.0%-93.3%, AUC = 0.780-0.982, p < 0.001). However, correcting the brain features for strength reduced the accuracy for predicting sex (accuracy = 57.3%-69.3%, AUC = 0.615-0.780). Here we demonstrate the effects of sex-correlated confounds in brain-based predictive models across multiple brain MRI modalities for both regression and classification models. We discuss implications of confounds in predictive modeling and the development of brain-based MRI biomarkers, as well as possible strategies to overcome these barriers.

Deep learning identifies morphological patterns of homologous recombination deficiency in luminal breast cancers from whole slide images

Homologous recombination DNA-repair deficiency (HRD) is becoming a well-recognized marker of platinum salt and polyADP-ribose polymerase inhibitor chemotherapies in ovarian and breast cancers. While large-scale screening for HRD using genomic markers is logistically and economically challenging, stained tissue slides are routinely acquired in clinical practice. With the objectives of providing a robust deep-learning method for HRD prediction from tissue slides and identifying related morphological phenotypes, we first show that digital pathology workflows are sensitive to potential biases in the training set, then we propose a method to overcome the influence of these biases, and we develop an interpretation method capable of identifying complex phenotypes. Application to our carefully curated in-house dataset allows us to predict HRD with high accuracy (area under the receiver-operator characteristics curve 0.86) and to identify morphological phenotypes related to HRD. In particular, the presence of laminated fibrosis and clear tumor cells associated with HRD open new hypotheses regarding its phenotypic impact.

Modern views of machine learning for precision psychiatry

In light of the National Institute of Mental Health (NIMH)'s Research Domain Criteria (RDoC), the advent of functional neuroimaging, novel technologies and methods provide new opportunities to develop precise and personalized prognosis and diagnosis of mental disorders. Machine learning (ML) and artificial intelligence (AI) technologies are playing an increasingly critical role in the new era of precision psychiatry. Combining ML/AI with neuromodulation technologies can potentially provide explainable solutions in clinical practice and effective therapeutic treatment. Advanced wearable and mobile technologies also call for the new role of ML/AI for digital phenotyping in mobile mental health. In this review, we provide a comprehensive review of ML methodologies and applications by combining neuroimaging, neuromodulation, and advanced mobile technologies in psychiatry practice. We further review the role of ML in molecular phenotyping and cross-species biomarker identification in precision psychiatry. We also discuss explainable AI (XAI) and neuromodulation in a closed human-in-the-loop manner and highlight the ML potential in multi-media information extraction and multi-modal data fusion. Finally, we discuss conceptual and practical challenges in precision psychiatry and highlight ML opportunities in future research.

Site effects how-to and when: An overview of retrospective techniques to accommodate site effects in multi-site neuroimaging analyses

Site differences, or systematic differences in feature distributions across multiple data-acquisition sites, are a known source of heterogeneity that may adversely affect large-scale meta- and mega-analyses of independently collected neuroimaging data. They influence nearly all multi-site imaging modalities and biomarkers, and methods to compensate for them can improve reliability and generalizability in the analysis of genetics, omics, and clinical data. The origins of statistical site effects are complex and involve both technical differences (scanner vendor, head coil, acquisition parameters, imaging processing) and differences in sample characteristics (inclusion/exclusion criteria, sample size, ancestry) between sites. In an age of expanding international consortium research, there is a growing need to disentangle technical site effects from sample characteristics of interest. Numerous statistical and machine learning methods have been developed to control for, model, or attenuate site effects - yet to date, no comprehensive review has discussed the benefits and drawbacks of each for different use cases. Here, we provide an overview of the different existing statistical and machine learning methods developed to remove unwanted site effects from independently collected neuroimaging samples. We focus on linear mixed effect models, the ComBat technique and its variants, adjustments based on image quality metrics, normative modeling, and deep learning approaches such as generative adversarial networks. For each method, we outline the statistical foundation and summarize strengths and weaknesses, including their assumptions and conditions of use. We provide information on software availability and comment on the ease of use and the applicability of these methods to different types of data. We discuss validation and comparative reports, mention caveats and provide guidance on when to use each method, depending on context and specific research questions.

Influence of contrast and texture based image modifications on the performance and attention shift of U-Net models for brain tissue segmentation

Contrast and texture modifications applied during training or test-time have recently shown promising results to enhance the generalization performance of deep learning segmentation methods in medical image analysis. However, a deeper understanding of this phenomenon has not been investigated. In this study, we investigated this phenomenon using a controlled experimental setting, using datasets from the Human Connectome Project and a large set of simulated MR protocols, in order to mitigate data confounders and investigate possible explanations as to why model performance changes when applying different levels of contrast and texture-based modifications. Our experiments confirm previous findings regarding the improved performance of models subjected to contrast and texture modifications employed during training and/or testing time, but further show the interplay when these operations are combined, as well as the regimes of model improvement/worsening across scanning parameters. Furthermore, our findings demonstrate a spatial attention shift phenomenon of trained models, occurring for different levels of model performance, and varying in relation to the type of applied image modification.

Artificial intelligence in medico-dental diagnostics of the face: a narrative review of opportunities and challenges

Objectives

This review aims to share the current developments of artificial intelligence (AI) solutions in the field of medico-dental diagnostics of the face. The primary focus of this review is to present the applicability of artificial neural networks (ANN) to interpret medical images, together with the associated opportunities, obstacles, and ethico-legal concerns.

Material and methods

Narrative literature review.

Results

Narrative literature review.

Conclusion

Curated facial images are widely available and easily accessible and are as such particularly suitable big data for ANN training. New AI solutions have the potential to change contemporary dentistry by optimizing existing processes and enriching dental care with the introduction of new tools for assessment or treatment planning. The analyses of health-related big data may also contribute to revolutionize personalized medicine through the detection of previously unknown associations. In regard to facial images, advances in medico-dental AI-based diagnostics include software solutions for the detection and classification of pathologies, for rating attractiveness and for the prediction of age or gender. In order for an ANN to be suitable for medical diagnostics of the face, the arising challenges regarding computation and management of the software are discussed, with special emphasis on the use of non-medical big data for ANN training. The legal and ethical ramifications of feeding patients’ facial images to a neural network for diagnostic purposes are related to patient consent, data privacy, data security, liability, and intellectual property. Current ethico-legal regulation practices seem incapable of addressing all concerns and ensuring accountability.

Clinical significance

While this review confirms the many benefits derived from AI solutions used for the diagnosis of medical images, it highlights the evident lack of regulatory oversight, the urgent need to establish licensing protocols, and the imperative to investigate the moral quality of new norms set with the implementation of AI applications in medico-dental diagnostics.

Identification of key somatic oncogenic mutation based on a confounder-free causal inference model

Abnormal cell proliferation and epithelial-mesenchymal transition (EMT) are the essential events that induce cancer initiation and progression. A fundamental goal in cancer research is to develop an efficient method to detect mutational genes capable of driving cancer. Although several computational methods have been proposed to identify these key mutations, many of them focus on the association between genetic mutations and functional changes in relevant biological processes, but not their real causality. Causal effect inference provides a way to estimate the real induce effect of a certain mutation on vital biological processes of cancer initiation and progression, through addressing the confounder bias due to neutral mutations and unobserved latent variables. In this study, integrating genomic and transcriptomic data, we construct a novel causal inference model based on a deep variational autoencoder to identify key oncogenic somatic mutations. Applied to 10 cancer types, our method quantifies the causal effect of genetic mutations on cell proliferation and EMT by reducing both observed and unobserved confounding biases. The experimental results indicate that genes with higher mutation frequency do not necessarily mean they are more potent in inducing cancer and promoting cancer development. Moreover, our study fills a gap in the use of machine learning for causal inference to identify oncogenic mutations.

Identifying key mutations of cancers is helpful to better understand the mechanisms of cancer cell transformation and is critical for therapeutic approaches. Besides sequence and structure-based computational approaches, some functional impact-based methods which consider the association between mutation events and the activity of cancer-related biological processes have also been developed to detect key mutations. However, these methods mainly consider the correlation but ignore that the correlation is far from causality due to the existence of observed and unobserved confounding factors. We develop a confounder-free machine learning-based causal inference framework to estimate the causal effect of mutations on abnormal cell proliferation and epithelial-mesenchymal transition (EMT). It fills a gap in the use of causal mechanisms to discover potential driver mutations in cancer biological systems. Applying our method to 10 cancer types, the identified key mutations are highly consistent with public well-verified ones. Additionally, some new key mutations have also been discovered.

A Penalty Approach for Normalizing Feature Distributions to Build Confounder-Free Models

Translating the use of modern machine learning algorithms into clinical applications requires settling challenges related to explain-ability and management of nuanced confounding factors. To suitably interpret the results, removing or explaining the effect of confounding variables (or metadata) is essential. Confounding variables affect the relationship between input training data and target outputs. Accordingly, when we train a model on such data, confounding variables will bias the distribution of the learned features. A recent promising solution, Meta-Data Normalization (MDN), estimates the linear relationship between the metadata and each feature based on a non-trainable closed-form solution. However, this estimation is confined by the sample size of a mini-batch and thereby may result in an oscillating performance. In this paper, we extend the MDN method by applying a Penalty approach (referred to as PDMN). We cast the problem into a bi-level nested optimization problem. We then approximate that objective using a penalty method so that the linear parameters within the MDN layer are trainable and learned on all samples. This enables PMDN to be plugged into any architectures, even those unfit to run batch-level operations such as transformers and recurrent models. We show improvement in model accuracy and independence from the confounders using PMDN over MDN in a synthetic experiment and a multi-label, multi-site classification of magnetic resonance images.

Conditional generation of medical time series for extrapolation to underrepresented populations

The widespread adoption of electronic health records (EHRs) and subsequent increased availability of longitudinal healthcare data has led to significant advances in our understanding of health and disease with direct and immediate impact on the development of new diagnostics and therapeutic treatment options. However, access to EHRs is often restricted due to their perceived sensitive nature and associated legal concerns, and the cohorts therein typically are those seen at a specific hospital or network of hospitals and therefore not representative of the wider population of patients. Here, we present HealthGen, a new approach for the conditional generation of synthetic EHRs that maintains an accurate representation of real patient characteristics, temporal information and missingness patterns. We demonstrate experimentally that HealthGen generates synthetic cohorts that are significantly more faithful to real patient EHRs than the current state-of-the-art, and that augmenting real data sets with conditionally generated cohorts of underrepresented subpopulations of patients can significantly enhance the generalisability of models derived from these data sets to different patient populations. Synthetic conditionally generated EHRs could help increase the accessibility of longitudinal healthcare data sets and improve the generalisability of inferences made from these data sets to underrepresented populations.

Artificial intelligence in differentiating tropical infections: A step ahead

Background and objective

Differentiating tropical infections are difficult due to its homogenous nature of clinical and laboratorial presentations among them. Sophisticated differential tests and prediction tools are better ways to tackle this issue. Here, we aimed to develop a clinician assisted decision making tool to differentiate the common tropical infections.

Methodology

A cross sectional study through 9 item self-administered questionnaire were performed to understand the need of developing a decision making tool and its parameters. The most significant differential parameters among the identified infections were measured through a retrospective study and decision tree was developed. Based on the parameters identified, a multinomial logistic regression model and a machine learning model were developed which could better differentiate the infection.

Results

A total of 40 physicians involved in the management of tropical infections were included for need analysis. Dengue, malaria, leptospirosis and scrub typhus were the common tropical infections in our settings. Sodium, total bilirubin, albumin, lymphocytes and platelets were the laboratory parameters; and abdominal pain, arthralgia, myalgia and urine output were the clinical presentation identified as better predictors. In multinomial logistic regression analysis with dengue as a reference revealed a predictability of 60.7%, 62.5% and 66% for dengue, malaria and leptospirosis, respectively, whereas, scrub typhus showed only 38% of predictability. The multi classification machine learning model observed to have an overall predictability of 55–60%, whereas a binary classification machine learning algorithms showed an average of 79–84% for one vs other and 69–88% for one vs one disease category.

Conclusion

This is a first of its kind study where both statistical and machine learning approaches were explored simultaneously for differentiating tropical infections. Machine learning techniques in healthcare sectors will aid in early detection and better patient care.

Distinguishing tropical infections is difficult due to its homogeneous nature from clinical and laboratory presentations among them. This is a first of its kind study where both statistical and machine learning approaches were explored simultaneously for differentiating tropical infections. Dengue, malaria, leptospirosis and scrub typhus were the common tropical infections in our settings as per the need analysis. Better predictors in terms of laboratory parameters and clinical presentations were identified from retrospective analysis and used for the regression and machine learning models. The parameters such as accuracy, true positive rate/sensitivity/recall, false positive rate, precision/positive predictive value, F-measure and ROC area for both the training and validation sets (10-fold cross validation) for all modelling approaches and diseases (One vs One and One vs others) were calculated. All the models observed to have an acceptable range of model performance in differentiating tropical infections. Albumin can be considered as the main parameter in differentiating these tropical infections. These models should be implemented in daily clinical routine practice via mobile or desktop assisted applications or tools.

Hierarchical confounder discovery in the experiment-machine learning cycle

The promise of machine learning (ML) to extract insights from high-dimensional datasets is tempered by confounding variables. It behooves scientists to determine if a model has extracted the desired information or instead fallen prey to bias. Due to features of natural phenomena and experimental design constraints, bioscience datasets are often organized in nested hierarchies that obfuscate the origins of confounding effects and render confounder amelioration methods ineffective. We propose a non-parametric statistical method called the rank-to-group (RTG) score that identifies hierarchical confounder effects in raw data and ML-derived embeddings. We show that RTG scores correctly assign the effects of hierarchical confounders when linear methods fail. In a public biomedical image dataset, we discover unreported effects of experimental design. We then use RTG scores to discover crossmodal correlated variability in a multi-phenotypic biological dataset. This approach should be generally useful in experiment-analysis cycles and to ensure confounder robustness in ML models.

•

Hierarchical confounder effects in complex datasets can bias ML models

•

RTG scoring identifies confounding variables in real-world biomedical data

•

RTG scoring can be used to guide model debiasing and inform experimental design

The promise of using machine learning (ML) to extract insights from high-dimensional data is tempered by the frequent presence of confounding variables. For example, models attempting to identify biomarkers of disease can be severely biased by disease-irrelevant features, such as the physical site where an experiment is performed. While we have many tools to grapple with known confounders, we lack a general method to identify which of a set of potential confounders warrant debiasing. Here, we present a simple non-parametric statistical method called the rank-to-group (RTG) score, which identifies hierarchical confounder effects in raw data and ML-derived data embeddings. We show that RTG scoring identifies previously unreported effects of experimental design in a public dataset and uncovers cross-model correlated variability in a multi-phenotypic biological dataset. This approach should be of general use in experiment-analysis cycles and to ensure confounder robustness in ML models.

How to remove or control confounds in predictive models, with applications to brain biomarkers

BACKGROUND

With increasing data sizes and more easily available computational methods, neurosciences rely more and more on predictive modeling with machine learning, e.g., to extract disease biomarkers. Yet, a successful prediction may capture a confounding effect correlated with the outcome instead of brain features specific to the outcome of interest. For instance, because patients tend to move more in the scanner than controls, imaging biomarkers of a disease condition may mostly reflect head motion, leading to inefficient use of resources and wrong interpretation of the biomarkers.

RESULTS

Here we study how to adapt statistical methods that control for confounds to predictive modeling settings. We review how to train predictors that are not driven by such spurious effects. We also show how to measure the unbiased predictive accuracy of these biomarkers, based on a confounded dataset. For this purpose, cross-validation must be modified to account for the nuisance effect. To guide understanding and practical recommendations, we apply various strategies to assess predictive models in the presence of confounds on simulated data and population brain imaging settings. Theoretical and empirical studies show that deconfounding should not be applied to the train and test data jointly: modeling the effect of confounds, on the training data only, should instead be decoupled from removing confounds.

CONCLUSIONS

Cross-validation that isolates nuisance effects gives an additional piece of information: confound-free prediction accuracy.

A Comprehensive Study on Bias in Artificial Intelligence Systems

Humans are social beings. Emotions, like their thoughts, play an essential role in decision-making. Today, artificial intelligence (AI) raises expectations for faster, more accurate, more rational, and fairer decisions with technological advancements. As a result, AI systems have often been seen as an ideal decision-making mechanism. But what if these systems decide against you based on gender, race, or other characteristics? Biased or unbiased AI, that's the question! The motivation of this study is to raise awareness among researchers about bias in AI and contribute to the advancement of AI studies and systems. As the primary purpose of this study is to examine bias in the decision-making process of AI systems, this paper focused on (1) bias in humans and AI, (2) the factors that lead to bias in AI systems, (3) current examples of bias in AI systems, and (4) various methods and recommendations to mitigate bias in AI systems.

A Novel Explainability Approach for Technology-Driven Translational Research on Brain Aging

Brain aging leads to difficulties in functional independence. Mitigating these difficulties can benefit from technology that predicts, monitors, and modifies brain aging. Translational research prioritizes solutions that can be causally linked to specific pathophysiologies at the same time as demonstrating improvements in impactful real-world outcome measures. This poses a challenge for brain aging technology that needs to address the tension between mechanism-driven precision and clinical relevance. In the current opinion, by synthesizing emerging mechanistic, translational, and clinical research-related frameworks, and our own development of technology-driven brain aging research, we suggest incorporating the appreciation of four desiderata (causality, informativeness, transferability, and fairness) of explainability into early-stage research that designs and tests brain aging technology. We apply a series of work on electrocardiography-based "peripheral" neuroplasticity markers from our work as an illustration of our proposed approach. We believe this novel approach will promote the development and adoption of brain aging technology that links and addresses brain pathophysiology and functional independence in the field of translational research.

Multi-label, multi-domain learning identifies compounding effects of HIV and cognitive impairment

Older individuals infected by Human Immunodeficiency Virus (HIV) are at risk for developing HIV-Associated Neurocognitive Disorder (HAND), i.e., from reduced cognitive functioning similar to HIV-negative individuals with Mild Cognitive Impairment (MCI) or to Alzheimer's Disease (AD) if more severely affected. Incompletely understood is how brain structure can serve to differentiate cognitive impairment (CI) in the HIV-positive (i.e., HAND) from the HIV-negative cohort (i.e., MCI and AD). To that end, we designed a multi-label classifier that labels the structural magnetic resonance images (MRI) of individuals by their HIV and CI status via two binary variables. Proper training of such an approach traditionally requires well-curated datasets containing large number of samples for each of the corresponding four cohorts (healthy controls, CI HIV-negative adults a.k.a. CI-only, HIV-positive patients without CI a.k.a. HIV-only, and HAND). Because of the rarity of such datasets, we proposed to improve training of the multi-label classifier via a multi-domain learning scheme that also incorporates domain-specific classifiers on auxiliary single-label datasets specific to either binary label. Specifically, we complement the training dataset of MRIs of the four cohorts (Control: 156, CI-only: 335, HIV-only: 37, HAND: 145) acquired by the Memory and Aging Center at the University of California - San Francisco with a CI-specific dataset only containing MRIs of HIV-negative subjects (Controls: 229, CI-only: 397) from the Alzheimer's Disease Neuroimaging Initiative and an HIV-specific dataset (Controls: 75, HIV-only: 75) provided by SRI International. Based on cross-validation on the UCSF dataset, the multi-domain and multi-label learning strategy leads to superior classification accuracy compared with one-domain or multi-class learning approaches, specifically for the undersampled HIV-only cohort. The 'prediction logits' of CI computed by the multi-label formulation also successfully stratify motor performance among the HIV-positive subjects (including HAND). Finally, brain patterns driving the subject-level predictions across all four cohorts characterize the independent and compounding effects of HIV and CI in the HAND cohort.

Co-optimization Learning Network for MRI Segmentation of Ischemic Penumbra Tissues

Convolutional neural networks (CNNs) have brought hope for the medical image auxiliary diagnosis. However, the shortfall of labeled medical image data is the bottleneck that limits the performance improvement of supervised CNN methods. In addition, annotating a large number of labeled medical image data is often expensive and time-consuming. In this study, we propose a co-optimization learning network (COL-Net) for Magnetic Resonance Imaging (MRI) segmentation of ischemic penumbra tissues. COL-Net base on the limited labeled samples and consists of an unsupervised reconstruction network (R), a supervised segmentation network (S), and a transfer block (T). The reconstruction network extracts the robust features from reconstructing pseudo unlabeled samples, which is the auxiliary branch of the segmentation network. The segmentation network is used to segment the target lesions under the limited labeled samples and the auxiliary of the reconstruction network. The transfer block is used to co-optimization the feature maps between the bottlenecks of the reconstruction network and segmentation network. We propose a mix loss function to optimize COL-Net. COL-Net is verified on the public ischemic penumbra segmentation challenge (SPES) with two dozen labeled samples. Results demonstrate that COL-Net has high predictive accuracy and generalization with the Dice coefficient of 0.79. The extended experiment also shows COL-Net outperforms most supervised segmentation methods. COL-Net is a meaningful attempt to alleviate the limited labeled sample problem in medical image segmentation.

Public Covid-19 X-ray datasets and their impact on model bias - A systematic review of a significant problem

Computer-aided-diagnosis and stratification of COVID-19 based on chest X-ray suffers from weak bias assessment and limited quality-control. Undetected bias induced by inappropriate use of datasets, and improper consideration of confounders prevents the translation of prediction models into clinical practice. By adopting established tools for model evaluation to the task of evaluating datasets, this study provides a systematic appraisal of publicly available COVID-19 chest X-ray datasets, determining their potential use and evaluating potential sources of bias. Only 9 out of more than a hundred identified datasets met at least the criteria for proper assessment of risk of bias and could be analysed in detail. Remarkably most of the datasets utilised in 201 papers published in peer-reviewed journals, are not among these 9 datasets, thus leading to models with high risk of bias. This raises concerns about the suitability of such models for clinical use. This systematic review highlights the limited description of datasets employed for modelling and aids researchers to select the most suitable datasets for their task.

xDEEP-MSI: Explainable Bias-Rejecting Microsatellite Instability Deep Learning System in Colorectal Cancer

The prediction of microsatellite instability (MSI) using deep learning (DL) techniques could have significant benefits, including reducing cost and increasing MSI testing of colorectal cancer (CRC) patients. Nonetheless, batch effects or systematic biases are not well characterized in digital histology models and lead to overoptimistic estimates of model performance. Methods to not only palliate but to directly abrogate biases are needed. We present a multiple bias rejecting DL system based on adversarial networks for the prediction of MSI in CRC from tissue microarrays (TMAs), trained and validated in 1788 patients from EPICOLON and HGUA. The system consists of an end-to-end image preprocessing module that tile samples at multiple magnifications and a tissue classification module linked to the bias-rejecting MSI predictor. We detected three biases associated with the learned representations of a baseline model: the project of origin of samples, the patient's spot and the TMA glass where each spot was placed. The system was trained to directly avoid learning the batch effects of those variables. The learned features from the bias-ablated model achieved maximum discriminative power with respect to the task and minimal statistical mean dependence with the biases. The impact of different magnifications, types of tissues and the model performance at tile vs patient level is analyzed. The AUC at tile level, and including all three selected tissues (tumor epithelium, mucin and lymphocytic regions) and 4 magnifications, was 0.87 ± 0.03 and increased to 0.9 ± 0.03 at patient level. To the best of our knowledge, this is the first work that incorporates a multiple bias ablation technique at the DL architecture in digital pathology, and the first using TMAs for the MSI prediction task.