Uncertainty-informed deep learning models enable high-confidence predictions for digital histopathology

Reliable Radiologic Skeletal Muscle Area Assessment - A Biomarker for Cancer Cachexia Diagnosis

Cancer cachexia is a common metabolic disorder characterized by severe muscle atrophy which is associated with poor prognosis and quality of life. Monitoring skeletal muscle area (SMA) longitudinally through computed tomography (CT) scans, an imaging modality routinely acquired in cancer care, is an effective way to identify and track this condition. However, existing tools often lack full automation and exhibit inconsistent accuracy, limiting their potential for integration into clinical workflows. To address these challenges, we developed SMAART-AI (Skeletal Muscle Assessment-Automated and Reliable Tool-based on AI), an end-to-end automated pipeline powered by deep learning models (nnU-Net 2D) trained on mid-third lumbar level CT images with 5-fold cross-validation, ensuring generalizability and robustness. SMAART-AI incorporates an uncertainty-based mechanism to flag high-error SMA predictions for expert review, enhancing reliability. We combined the SMA, skeletal muscle index, BMI, and clinical data to train a multi-layer perceptron (MLP) model designed to predict cachexia at the time of cancer diagnosis. Tested on the gastroesophageal cancer dataset, SMAART-AI achieved a Dice score of 97.80% ± 0.93%, with SMA estimated across all four datasets in this study at a median absolute error of 2.48% compared to manual annotations with SliceOmatic. Uncertainty metrics-variance, entropy, and coefficient of variation-strongly correlated with SMA prediction errors (0.83, 0.76, and 0.73 respectively). The MLP model predicts cachexia with 79% precision, providing clinicians with a reliable tool for early diagnosis and intervention. By combining automation, accuracy, and uncertainty awareness, SMAART-AI bridges the gap between research and clinical application, offering a transformative approach to managing cancer cachexia.

Self-interactive learning: Fusion and evolution of multi-scale histomorphology features for molecular traits prediction in computational pathology

Predicting disease-related molecular traits from histomorphology brings great opportunities for precision medicine. Despite the rich information present in histopathological images, extracting fine-grained molecular features from standard whole slide images (WSI) is non-trivial. The task is further complicated by the lack of annotations for subtyping and contextual histomorphological features that might span multiple scales. This work proposes a novel multiple-instance learning (MIL) framework capable of WSI-based cancer morpho-molecular subtyping by fusion of different-scale features. Our method, debuting as Inter-MIL, follows a weakly-supervised scheme. It enables the training of the patch-level encoder for WSI in a task-aware optimisation procedure, a step normally not modelled in most existing MIL-based WSI analysis frameworks. We demonstrate that optimising the patch-level encoder is crucial to achieving high-quality fine-grained and tissue-level subtyping results and offers a significant improvement over task-agnostic encoders. Our approach deploys a pseudo-label propagation strategy to update the patch encoder iteratively, allowing discriminative subtype features to be learned. This mechanism also empowers extracting fine-grained attention within image tiles (the small patches), a task largely ignored in most existing weakly supervised-based frameworks. With Inter-MIL, we carried out four challenging cancer molecular subtyping tasks in the context of ovarian, colorectal, lung, and breast cancer. Extensive evaluation results show that Inter-MIL is a robust framework for cancer morpho-molecular subtyping with superior performance compared to several recently proposed methods, in small dataset scenarios where the number of available training slides is less than 100. The iterative optimisation mechanism of Inter-MIL significantly improves the quality of the image features learned by the patch embedded and generally directs the attention map to areas that better align with experts' interpretation, leading to the identification of more reliable histopathology biomarkers. Moreover, an external validation cohort is used to verify the robustness of Inter-MIL on molecular trait prediction.

Implementing Trust in Non-Small Cell Lung Cancer Diagnosis with a Conformalized Uncertainty-Aware AI Framework in Whole-Slide Images

Ensuring trustworthiness is fundamental to the development of artificial intelligence (AI) that is considered societally responsible, particularly in cancer diagnostics, where a misdiagnosis can have dire consequences. Current digital pathology AI models lack systematic solutions to address trustworthiness concerns arising from model limitations and data discrepancies between model deployment and development environments. To address this issue, we developed TRUECAM, a framework designed to ensure both data and model trustworthiness in non-small cell lung cancer subtyping with whole-slide images. TRUECAM integrates 1) a spectral-normalized neural Gaussian process for identifying out-of-scope inputs and 2) an ambiguity-guided elimination of tiles to filter out highly ambiguous regions, addressing data trustworthiness, as well as 3) conformal prediction to ensure controlled error rates. We systematically evaluated the framework across multiple large-scale cancer datasets, leveraging both task-specific and foundation models, illustrate that an AI model wrapped with TRUECAM significantly outperforms models that lack such guidance, in terms of classification accuracy, robustness, interpretability, and data efficiency, while also achieving improvements in fairness. These findings highlight TRUECAM as a versatile wrapper framework for digital pathology AI models with diverse architectural designs, promoting their responsible and effective applications in real-world settings.

Closing the Automation Gap: Robust AI for Dual-Stain Cervical Cancer Screening Triage

Dual-stain cytology, using p16 and Ki67, is superior to conventional PAP cytology for triage of HPV-positive test results in cervical cancer screening. Its AI-based evaluation can remove subjectivity, improve performance and facilitate implementation. Using 5,722 dual-stain slides from population-based screening cohorts, we developed and validated Cytoreader-V2. In the SurePath Kaiser Implementation Study, Cytoreader-V2 achieved 87.2%/57.8% (sensitivity/specificity) compared to 89.9/52.6 (manual DS) and 85.8/41.9 (Pap cytology). In the Thin-Prep Biopsy Study, it reached 95.7/44.4 versus 89.4/35.0 (manual DS), and in anal DS cytology slides, 87.0/41.3 compared to 87.0/27.7 (manual). Robustness testing demonstrated significant stability across image transformations. Cytoreader-V2 improves specificity and reproducibility compared to manual dual-stain reading while maintaining high sensitivity. Its adaptability across populations with consistent performance makes it scalable for diverse clinical settings. Cytoreader-V2 can be a transformative tool in global cervical cancer screening as a critical AI applications in digital pathology.

Integrating artificial intelligence with smartphone-based imaging for cancer detection in vivo

Cancer is a major global health challenge, accounting for nearly one in six deaths worldwide. Early diagnosis significantly improves survival rates and patient outcomes, yet in resource-limited settings, the scarcity of medical resources often leads to late-stage diagnosis. Integrating artificial intelligence (AI) with smartphone-based imaging systems offers a promising solution by providing portable, cost-effective, and widely accessible tools for early cancer detection. This paper introduces advanced smartphone-based imaging systems that utilize various imaging modalities for in vivo detection of different cancer types and highlights the advancements of AI for in vivo cancer detection in smartphone-based imaging. However, these compact smartphone systems face challenges like low imaging quality and restricted computing power. The use of advanced AI algorithms to address the optical and computational limitations of smartphone-based imaging systems provides promising solutions. AI-based cancer detection also faces challenges. Transparency and reliability are critical factors in gaining the trust and acceptance of AI algorithms for clinical application, explainable and uncertainty-aware AI breaks the black box and will shape the future AI development in early cancer detection. The challenges and solutions for improving AI accuracy, transparency, and reliability are general issues in AI applications, the AI technologies, limitations, and potentials discussed in this paper are applicable to a wide range of biomedical imaging diagnostics beyond smartphones or cancer-specific applications. Smartphone-based multimodal imaging systems and deep learning algorithms for multimodal data analysis are also growing trends, as this approach can provide comprehensive information about the tissue being examined. Future opportunities and perspectives of AI-integrated smartphone imaging systems will be to make cutting-edge diagnostic tools more affordable and accessible, ultimately enabling early cancer detection for a broader population.

Incorporating label uncertainty during the training of convolutional neural networks improves performance for the discrimination between certain and inconclusive cases in dopamine transporter SPECT

PURPOSE

Deep convolutional neural networks (CNN) hold promise for assisting the interpretation of dopamine transporter (DAT)-SPECT. For improved communication of uncertainty to the user it is crucial to reliably discriminate certain from inconclusive cases that might be misclassified by strict application of a predefined decision threshold on the CNN output. This study tested two methods to incorporate existing label uncertainty during the training to improve the utility of the CNN sigmoid output for this task.

METHODS

Three datasets were used retrospectively: a "development" dataset (n = 1740) for CNN training, validation and testing, two independent out-of-distribution datasets (n = 640, 645) for testing only. In the development dataset, binary classification based on visual inspection was performed carefully by three well-trained readers. A ResNet-18 architecture was trained for binary classification of DAT-SPECT using either a randomly selected vote ("random vote training", RVT), the proportion of "reduced" votes ( "average vote training", AVT) or the majority vote (MVT) across the three readers as reference standard. Balanced accuracy was computed separately for "inconclusive" sigmoid outputs (within a predefined interval around the 0.5 decision threshold) and for "certain" (non-inconclusive) sigmoid outputs.

RESULTS

The proportion of "inconclusive" test cases that had to be accepted to achieve a given balanced accuracy in the "certain" test case was lower with RVT and AVT than with MVT in all datasets (e.g., 1.9% and 1.2% versus 2.8% for 98% balanced accuracy in "certain" test cases from the development dataset). In addition, RVT and AVT resulted in slightly higher balanced accuracy in all test cases independent of their certainty (97.3% and 97.5% versus 97.0% in the development dataset).

CONCLUSION

Making between-readers-discrepancy known to CNN during the training improves the utility of their sigmoid output to discriminate certain from inconclusive cases that might be misclassified by the CNN when the predefined decision threshold is strictly applied. This does not compromise on overall accuracy.

Reducing inference cost of Alzheimer's disease identification using an uncertainty-aware ensemble of uni-modal and multi-modal learners

While multi-modal deep learning approaches trained using magnetic resonance imaging (MRI) and fluorodeoxyglucose positron emission tomography (FDG PET) data have shown promise in the accurate identification of Alzheimer's disease, their clinical applicability is hindered by the assumption that both modalities are always available during model inference. In practice, clinicians adjust diagnostic tests based on available information and specific clinical contexts. We propose a novel MRI- and FDG PET-based multi-modal deep learning approach that mimics clinical decision-making by incorporating uncertainty estimates of an MRI-based model (generated using Monte Carlo dropout and evidential deep learning) to determine the necessity of an FDG PET scan, and only inputting the FDG PET to a multi-modal model when required. This approach significantly reduces the reliance on FDG PET scans, which are costly and expose patients to radiation. Our approach reduces the need for FDG PET by up to 92% without compromising model performance, thus optimizing resource use and patient safety. Furthermore, using a global model explanation technique, we provide insights into how anatomical changes in brain regions-such as the entorhinal cortex, amygdala, and ventricles-can positively or negatively influence the need for FDG PET scans in alignment with clinical understanding of Alzheimer's disease.

Uncertainty-aware regression model to predict post-operative visual acuity in patients with macular holes

Full-thickness macular holes are a relatively common and visually disabling condition with a prevalence of approximately 0.5% in the over-40-year-old age group. If left untreated, the hole typically enlarges, reducing visual acuity (VA) below the definition of blindness in the eye affected. They are now routinely treated with surgery, which can close the hole and improve vision in most cases. The extent of improvement, however, is variable and dependent on the size of the hole and other features which can be discerned in spectral-domain optical coherence tomography imaging, which is now routinely available in eye clinics globally. Artificial intelligence (AI) models have been developed to enable surgical decision-making and have achieved relatively high predictive performance. However, their black-box behavior is opaque to users and uncertainty associated with their predictions is not typically stated, leading to a lack of trust among clinicians and patients. In this paper, we describe an uncertainty-aware regression model (U-ARM) for predicting VA for people undergoing macular hole surgery using preoperative spectral-domain optical coherence tomography images, achieving an MAE of 6.07, RMSE of 9.11 and R2 of 0.47 in internal tests, and an MAE of 6.49, RMSE of 9.49, and R2 of 0.42 in external tests. In addition to predicting VA following surgery, U-ARM displays its associated uncertainty, a p-value of <0.005 in internal and external tests, showing the predictions are not due to random chance. We then qualitatively evaluated the performance of U-ARM. Lastly, we demonstrate out-of-sample data performance, generalizing well to data outside the training distribution, low-quality images, and unseen instances not encountered during training. The results show that U-ARM outperforms commonly used methods in terms of prediction and reliability. U-ARM is thus a promising approach for clinical settings and can improve the reliability of AI models in predicting VA.

FedDBL: Communication and Data Efficient Federated Deep-Broad Learning for Histopathological Tissue Classification

Histopathological tissue classification is a fundamental task in computational pathology. Deep learning (DL)-based models have achieved superior performance but centralized training suffers from the privacy leakage problem. Federated learning (FL) can safeguard privacy by keeping training samples locally, while existing FL-based frameworks require a large number of well-annotated training samples and numerous rounds of communication which hinder their viability in real-world clinical scenarios. In this article, we propose a lightweight and universal FL framework, named federated deep-broad learning (FedDBL), to achieve superior classification performance with limited training samples and only one-round communication. By simply integrating a pretrained DL feature extractor, a fast and lightweight broad learning inference system with a classical federated aggregation approach, FedDBL can dramatically reduce data dependency and improve communication efficiency. Five-fold cross-validation demonstrates that FedDBL greatly outperforms the competitors with only one-round communication and limited training samples, while it even achieves comparable performance with the ones under multiple-round communications. Furthermore, due to the lightweight design and one-round communication, FedDBL reduces the communication burden from 4.6 GB to only 138.4 KB per client using the ResNet-50 backbone at 50-round training. Extensive experiments also show the scalability of FedDBL on model generalization to the unseen dataset, various client numbers, model personalization and other image modalities. Since no data or deep model sharing across different clients, the privacy issue is well-solved and the model security is guaranteed with no model inversion attack risk. Code is available at https://github.com/tianpeng-deng/FedDBL.

Personality Science in the Digital Age: The Promises and Challenges of Psychological Targeting for Personalized Behavior-Change Interventions at Scale

With the rapidly growing availability of scalable psychological assessments, personality science holds great promise for the scientific study and applied use of customized behavior-change interventions. To facilitate this development, we propose a classification system that divides psychological targeting into two approaches that differ in the process by which interventions are designed: audience-to-content matching or content-to-audience matching. This system is both integrative and generative: It allows us to (a) integrate existing research on personalized interventions from different psychological subdisciplines (e.g., political, educational, organizational, consumer, and clinical and health psychology) and to (b) articulate open questions that generate promising new avenues for future research. Our objective is to infuse personality science into intervention research and encourage cross-disciplinary collaborations within and outside of psychology. To ensure the development of personality-customized interventions aligns with the broader interests of individuals (and society at large), we also address important ethical considerations for the use of psychological targeting (e.g., privacy, self-determination, and equity) and offer concrete guidelines for researchers and practitioners.

Enhancing a deep learning model for pulmonary nodule malignancy risk estimation in chest CT with uncertainty estimation

OBJECTIVE

To investigate the effect of uncertainty estimation on the performance of a Deep Learning (DL) algorithm for estimating malignancy risk of pulmonary nodules.

METHODS AND MATERIALS

In this retrospective study, we integrated an uncertainty estimation method into a previously developed DL algorithm for nodule malignancy risk estimation. Uncertainty thresholds were developed using CT data from the Danish Lung Cancer Screening Trial (DLCST), containing 883 nodules (65 malignant) collected between 2004 and 2010. We used thresholds on the 90th and 95th percentiles of the uncertainty score distribution to categorize nodules into certain and uncertain groups. External validation was performed on clinical CT data from a tertiary academic center containing 374 nodules (207 malignant) collected between 2004 and 2012. DL performance was measured using area under the ROC curve (AUC) for the full set of nodules, for the certain cases and for the uncertain cases. Additionally, nodule characteristics were compared to identify trends for inducing uncertainty.

RESULTS

The DL algorithm performed significantly worse in the uncertain group compared to the certain group of DLCST (AUC 0.62 (95% CI: 0.49, 0.76) vs 0.93 (95% CI: 0.88, 0.97); p < .001) and the clinical dataset (AUC 0.62 (95% CI: 0.50, 0.73) vs 0.90 (95% CI: 0.86, 0.94); p < .001). The uncertain group included larger benign nodules as well as more part-solid and non-solid nodules than the certain group.

CONCLUSION

The integrated uncertainty estimation showed excellent performance for identifying uncertain cases in which the DL-based nodule malignancy risk estimation algorithm had significantly worse performance.

CLINICAL RELEVANCE STATEMENT

Deep Learning algorithms often lack the ability to gauge and communicate uncertainty. For safe clinical implementation, uncertainty estimation is of pivotal importance to identify cases where the deep learning algorithm harbors doubt in its prediction.

KEY POINTS

• Deep learning (DL) algorithms often lack uncertainty estimation, which potentially reduce the risk of errors and improve safety during clinical adoption of the DL algorithm. • Uncertainty estimation identifies pulmonary nodules in which the discriminative performance of the DL algorithm is significantly worse. • Uncertainty estimation can further enhance the benefits of the DL algorithm and improve its safety and trustworthiness.

Artificial Intelligence Applications in Oral Cancer and Oral Dysplasia

Oral squamous cell carcinoma (OSCC) is a highly unpredictable disease with devastating mortality rates that have not changed over the past decades, in the face of advancements in treatments and biomarkers, which have improved survival for other cancers. Delays in diagnosis are frequent, leading to more disfiguring treatments and poor outcomes for patients. The clinical challenge lies in identifying those patients at the highest risk of developing OSCC. Oral epithelial dysplasia (OED) is a precursor of OSCC with highly variable behavior across patients. There is no reliable clinical, pathological, histological, or molecular biomarker to determine individual risk in OED patients. Similarly, there are no robust biomarkers to predict treatment outcomes or mortality in OSCC patients. This review aims to highlight advancements in artificial intelligence (AI)-based methods to develop predictive biomarkers of OED transformation to OSCC or predictive biomarkers of OSCC mortality and treatment response. Biomarkers such as S100A7 demonstrate promising appraisal for the risk of malignant transformation of OED. Machine learning-enhanced multiplex immunohistochemistry workflows examine immune cell patterns and organization within the tumor immune microenvironment to generate outcome predictions in immunotherapy. Deep learning (DL) is an AI-based method using an extended neural network or related architecture with multiple "hidden" layers of simulated neurons to combine simple visual features into complex patterns. DL-based digital pathology is currently being developed to assess OED and OSCC outcomes. The integration of machine learning in epigenomics aims to examine the epigenetic modification of diseases and improve our ability to detect, classify, and predict outcomes associated with epigenetic marks. Collectively, these tools showcase promising advancements in discovery and technology, which may provide a potential solution to addressing the current limitations in predicting OED transformation and OSCC behavior, both of which are clinical challenges that must be addressed in order to improve OSCC survival.

A review of uncertainty quantification in medical image analysis: Probabilistic and non-probabilistic methods

The comprehensive integration of machine learning healthcare models within clinical practice remains suboptimal, notwithstanding the proliferation of high-performing solutions reported in the literature. A predominant factor hindering widespread adoption pertains to an insufficiency of evidence affirming the reliability of the aforementioned models. Recently, uncertainty quantification methods have been proposed as a potential solution to quantify the reliability of machine learning models and thus increase the interpretability and acceptability of the results. In this review, we offer a comprehensive overview of the prevailing methods proposed to quantify the uncertainty inherent in machine learning models developed for various medical image tasks. Contrary to earlier reviews that exclusively focused on probabilistic methods, this review also explores non-probabilistic approaches, thereby furnishing a more holistic survey of research pertaining to uncertainty quantification for machine learning models. Analysis of medical images with the summary and discussion on medical applications and the corresponding uncertainty evaluation protocols are presented, which focus on the specific challenges of uncertainty in medical image analysis. We also highlight some potential future research work at the end. Generally, this review aims to allow researchers from both clinical and technical backgrounds to gain a quick and yet in-depth understanding of the research in uncertainty quantification for medical image analysis machine learning models.

E2-MIL: An explainable and evidential multiple instance learning framework for whole slide image classification

Multiple instance learning (MIL)-based methods have been widely adopted to process the whole slide image (WSI) in the field of computational pathology. Due to the sparse slide-level supervision, these methods usually lack good localization on the tumor regions, leading to poor interpretability. Moreover, they lack robust uncertainty estimation of prediction results, leading to poor reliability. To solve the above two limitations, we propose an explainable and evidential multiple instance learning (E2-MIL) framework for whole slide image classification. E2-MIL is mainly composed of three modules: a detail-aware attention distillation module (DAM), a structure-aware attention refined module (SRM), and an uncertainty-aware instance classifier (UIC). Specifically, DAM helps the global network locate more detail-aware positive instances by utilizing the complementary sub-bags to learn detailed attention knowledge from the local network. In addition, a masked self-guidance loss is also introduced to help bridge the gap between the slide-level labels and instance-level classification tasks. SRM generates a structure-aware attention map that locates the entire tumor region structure by effectively modeling the spatial relations between clustering instances. Moreover, UIC provides accurate instance-level classification results and robust predictive uncertainty estimation to improve the model reliability based on subjective logic theory. Extensive experiments on three large multi-center subtyping datasets demonstrate both slide-level and instance-level performance superiority of E2-MIL.

Next-generation lung cancer pathology: Development and validation of diagnostic and prognostic algorithms

Non-small cell lung cancer (NSCLC) is one of the most common malignant tumors. In this study, we develop a clinically useful computational pathology platform for NSCLC that can be a foundation for multiple downstream applications and provide immediate value for patient care optimization and individualization. We train the primary multi-class tissue segmentation algorithm on a substantial, high-quality, manually annotated dataset of whole-slide images with lung adenocarcinoma and squamous cell carcinomas. We investigate two downstream applications. NSCLC subtyping algorithm is trained and validated using a large, multi-institutional (n = 6), multi-scanner (n = 5), international cohort of NSCLC cases (slides/patients 4,097/1,527). Moreover, we develop four AI-derived, fully explainable, quantitative, prognostic parameters (based on tertiary lymphoid structure and necrosis assessment) and validate them for different clinical endpoints. The computational platform enables the high-precision, quantitative analysis of H&E-stained slides. The developed prognostic parameters facilitate robust and independent risk stratification of patients with NSCLC.

A deep learning-based algorithm for automatic detection of perilunate dislocation in frontal wrist radiographs

This study proposes a Deep Learning algorithm to automatically detect perilunate dislocation in anteroposterior wrist radiographs. A total of 374 annotated radiographs, 345 normal and 29 pathological, of skeletally mature adolescents and adults aged ≥16 years were used to train, validate and test two YOLOv8 deep neural models. The training set included 245 normal and 15 pathological radiographs; the pathological training set was supplemented by 240 radiographs obtained by data augmentation. The test set comprised 30 normal and 10 pathological radiographs. The first model was used for detecting the carpal region, and the second for segmenting a region between Gilula's 2nd and 3rd arcs. The output of the segmentation model, trained multiple times with varying random initial parameter values and augmentations, was then assigned a probability of being normal or pathological through ensemble averaging. In the study dataset, the algorithm achieved an overall F1-score of 0.880: 0.928 in the normal subgroup, with 1.0 precision, and 0.833 in the pathological subgroup, with 1.0 recall (or sensitivity), demonstrating that diagnosis of perilunate dislocation can be improved by automatic analysis of anteroposterior radiographs. LEVEL OF EVIDENCE: : III.

Developing a low-cost, open-source, locally manufactured workstation and computational pipeline for automated histopathology evaluation using deep learning

BACKGROUND

Deployment and access to state-of-the-art precision medicine technologies remains a fundamental challenge in providing equitable global cancer care in low-resource settings. The expansion of digital pathology in recent years and its potential interface with diagnostic artificial intelligence algorithms provides an opportunity to democratize access to personalized medicine. Current digital pathology workstations, however, cost thousands to hundreds of thousands of dollars. As cancer incidence rises in many low- and middle-income countries, the validation and implementation of low-cost automated diagnostic tools will be crucial to helping healthcare providers manage the growing burden of cancer.

METHODS

Here we describe a low-cost ($230) workstation for digital slide capture and computational analysis composed of open-source components. We analyze the predictive performance of deep learning models when they are used to evaluate pathology images captured using this open-source workstation versus images captured using common, significantly more expensive hardware. Validation studies assessed model performance on three distinct datasets and predictive models: head and neck squamous cell carcinoma (HPV positive versus HPV negative), lung cancer (adenocarcinoma versus squamous cell carcinoma), and breast cancer (invasive ductal carcinoma versus invasive lobular carcinoma).

FINDINGS

When compared to traditional pathology image capture methods, low-cost digital slide capture and analysis with the open-source workstation, including the low-cost microscope device, was associated with model performance of comparable accuracy for breast, lung, and HNSCC classification. At the patient level of analysis, AUROC was 0.84 for HNSCC HPV status prediction, 1.0 for lung cancer subtype prediction, and 0.80 for breast cancer classification.

INTERPRETATION

Our ability to maintain model performance despite decreased image quality and low-power computational hardware demonstrates that it is feasible to massively reduce costs associated with deploying deep learning models for digital pathology applications. Improving access to cutting-edge diagnostic tools may provide an avenue for reducing disparities in cancer care between high- and low-income regions.

FUNDING

Funding for this project including personnel support was provided via grants from NIH/NCIR25-CA240134, NIH/NCIU01-CA243075, NIH/NIDCRR56-DE030958, NIH/NCIR01-CA276652, NIH/NCIK08-CA283261, NIH/NCI-SOAR25CA240134, SU2C (Stand Up to Cancer) Fanconi Anemia Research Fund - Farrah Fawcett Foundation Head and Neck Cancer Research Team Grant, and the European UnionHorizon Program (I3LUNG).

Memristive Monte Carlo DropConnect crossbar array enabled by device and algorithm co-design

Device and algorithm co-design aims to develop energy-efficient hardware that directly implements complex algorithms and optimizes algorithms to match the hardware's characteristics. Specifically, neuromorphic computing algorithms are constantly growing in complexity, necessitating an ongoing search for hardware implementations capable of handling these intricate algorithms. Here, we present a memristive Monte Carlo DropConnect (MC-DC) crossbar array developed through a hardware algorithm co-design approach. To implement the MC-DC neural network, stochastic switching and analog memory characteristics are required, and we achieved them using Ag-based diffusive selectors and Ru-based electrochemical metalization (ECM) memristors, respectively. The devices were integrated with a one-selector one-memristor (1S1M) structure, and their well-matched operating voltages and currents enabled stochastic readout and deterministic analog programming. With the integrated hardware, we successfully demonstrated the MC-DC operation. Additionally, the selector allowed for the control of switching polarity, and by understanding this hardware characteristic, we were able to modify the algorithm to fit it and further improve the network performance.

Artificial intelligence and endo-histo-omics: new dimensions of precision endoscopy and histology in inflammatory bowel disease

Integrating artificial intelligence into inflammatory bowel disease (IBD) has the potential to revolutionise clinical practice and research. Artificial intelligence harnesses advanced algorithms to deliver accurate assessments of IBD endoscopy and histology, offering precise evaluations of disease activity, standardised scoring, and outcome prediction. Furthermore, artificial intelligence offers the potential for a holistic endo-histo-omics approach by interlacing and harmonising endoscopy, histology, and omics data towards precision medicine. The emerging applications of artificial intelligence could pave the way for personalised medicine in IBD, offering patient stratification for the most beneficial therapy with minimal risk. Although artificial intelligence holds promise, challenges remain, including data quality, standardisation, reproducibility, scarcity of randomised controlled trials, clinical implementation, ethical concerns, legal liability, and regulatory issues. The development of standardised guidelines and interdisciplinary collaboration, including policy makers and regulatory agencies, is crucial for addressing these challenges and advancing artificial intelligence in IBD clinical practice and trials.

Learning generalizable AI models for multi-center histopathology image classification

Investigation of histopathology slides by pathologists is an indispensable component of the routine diagnosis of cancer. Artificial intelligence (AI) has the potential to enhance diagnostic accuracy, improve efficiency, and patient outcomes in clinical pathology. However, variations in tissue preparation, staining protocols, and histopathology slide digitization could result in over-fitting of deep learning models when trained on the data from only one center, thereby underscoring the necessity to generalize deep learning networks for multi-center use. Several techniques, including the use of grayscale images, color normalization techniques, and Adversarial Domain Adaptation (ADA) have been suggested to generalize deep learning algorithms, but there are limitations to their effectiveness and discriminability. Convolutional Neural Networks (CNNs) exhibit higher sensitivity to variations in the amplitude spectrum, whereas humans predominantly rely on phase-related components for object recognition. As such, we propose Adversarial fourIer-based Domain Adaptation (AIDA) which applies the advantages of a Fourier transform in adversarial domain adaptation. We conducted a comprehensive examination of subtype classification tasks in four cancers, incorporating cases from multiple medical centers. Specifically, the datasets included multi-center data for 1113 ovarian cancer cases, 247 pleural cancer cases, 422 bladder cancer cases, and 482 breast cancer cases. Our proposed approach significantly improved performance, achieving superior classification results in the target domain, surpassing the baseline, color augmentation and normalization techniques, and ADA. Furthermore, extensive pathologist reviews suggested that our proposed approach, AIDA, successfully identifies known histotype-specific features. This superior performance highlights AIDA's potential in addressing generalization challenges in deep learning models for multi-center histopathology datasets.

ChemFH: an integrated tool for screening frequent false positives in chemical biology and drug discovery

High-throughput screening rapidly tests an extensive array of chemical compounds to identify hit compounds for specific biological targets in drug discovery. However, false-positive results disrupt hit compound screening, leading to wastage of time and resources. To address this, we propose ChemFH, an integrated online platform facilitating rapid virtual evaluation of potential false positives, including colloidal aggregators, spectroscopic interference compounds, firefly luciferase inhibitors, chemical reactive compounds, promiscuous compounds, and other assay interferences. By leveraging a dataset containing 823 391 compounds, we constructed high-quality prediction models using multi-task directed message-passing network (DMPNN) architectures combining uncertainty estimation, yielding an average AUC value of 0.91. Furthermore, ChemFH incorporated 1441 representative alert substructures derived from the collected data and ten commonly used frequent hitter screening rules. ChemFH was validated with an external set of 75 compounds. Subsequently, the virtual screening capability of ChemFH was successfully confirmed through its application to five virtual screening libraries. Furthermore, ChemFH underwent additional validation on two natural products and FDA-approved drugs, yielding reliable and accurate results. ChemFH is a comprehensive, reliable, and computationally efficient screening pipeline that facilitates the identification of true positive results in assays, contributing to enhanced efficiency and success rates in drug discovery. ChemFH is freely available via https://chemfh.scbdd.com/.

Advancements in computer vision and pathology: Unraveling the potential of artificial intelligence for precision diagnosis and beyond

The integration of computer vision into pathology through slide digitalization represents a transformative leap in the field's evolution. Traditional pathology methods, while reliable, are often time-consuming and susceptible to intra- and interobserver variability. In contrast, computer vision, empowered by artificial intelligence (AI) and machine learning (ML), promises revolutionary changes, offering consistent, reproducible, and objective results with ever-increasing speed and scalability. The applications of advanced algorithms and deep learning architectures like CNNs and U-Nets augment pathologists' diagnostic capabilities, opening new frontiers in automated image analysis. As these technologies mature and integrate into digital pathology workflows, they are poised to provide deeper insights into disease processes, quantify and standardize biomarkers, enhance patient outcomes, and automate routine tasks, reducing pathologists' workload. However, this transformative force calls for cross-disciplinary collaboration between pathologists, computer scientists, and industry innovators to drive research and development. While acknowledging its potential, this chapter addresses the limitations of AI in pathology, encompassing technical, practical, and ethical considerations during development and implementation.

Application of simultaneous uncertainty quantification and segmentation for oropharyngeal cancer use-case with Bayesian deep learning

BACKGROUND

Radiotherapy is a core treatment modality for oropharyngeal cancer (OPC), where the primary gross tumor volume (GTVp) is manually segmented with high interobserver variability. This calls for reliable and trustworthy automated tools in clinician workflow. Therefore, accurate uncertainty quantification and its downstream utilization is critical.

METHODS

Here we propose uncertainty-aware deep learning for OPC GTVp segmentation, and illustrate the utility of uncertainty in multiple applications. We examine two Bayesian deep learning (BDL) models and eight uncertainty measures, and utilize a large multi-institute dataset of 292 PET/CT scans to systematically analyze our approach.

RESULTS

We show that our uncertainty-based approach accurately predicts the quality of the deep learning segmentation in 86.6% of cases, identifies low performance cases for semi-automated correction, and visualizes regions of the scans where the segmentations likely fail.

CONCLUSIONS

Our BDL-based analysis provides a first-step towards more widespread implementation of uncertainty quantification in OPC GTVp segmentation.

A Human-Centered AI Framework for Efficient Labelling of ECGs From Drug Safety Trials

Drug safety trials require substantial ECG labelling like, in thorough QT studies, measurements of the QT interval, whose prolongation is a biomarker of proarrhythmic risk. The traditional method of manually measuring the QT interval is time-consuming and error-prone. Studies have demonstrated the potential of deep learning (DL)-based methods to automate this task but expert validation of these computerized measurements remains of paramount importance, particularly for abnormal ECG recordings. In this paper, we propose a highly automated framework that combines such a DL-based QT estimator with human expertise. The framework consists of 3 key components: (1) automated QT measurement with uncertainty quantification (2) expert review of a few DL-based measurements, mostly those with high model uncertainty and (3) recalibration of the unreviewed measurements based on the expert-validated data. We assess its effectiveness on 3 drug safety trials and show that it can significantly reduce effort required for ECG labelling-in our experiments only 10% of the data were reviewed per trial-while maintaining high levels of QT accuracy. Our study thus demonstrates the possibility of productive human-machine collaboration in ECG analysis without any compromise on the reliability of subsequent clinical interpretations.

Deep neural network uncertainty estimation for early oral cancer diagnosis

BACKGROUND

Early diagnosis in oral cancer is essential to reduce both morbidity and mortality. This study explores the use of uncertainty estimation in deep learning for early oral cancer diagnosis.

METHODS

We develop a Bayesian deep learning model termed 'Probabilistic HRNet', which utilizes the ensemble MC dropout method on HRNet. Additionally, two oral lesion datasets with distinct distributions are created. We conduct a retrospective study to assess the predictive performance and uncertainty of Probabilistic HRNet across these datasets.

RESULTS

Probabilistic HRNet performs optimally on the In-domain test set, achieving an F1 score of 95.3% and an AUC of 96.9% by excluding the top 30% high-uncertainty samples. For evaluations on the Domain-shift test set, the results show an F1 score of 64.9% and an AUC of 80.3%. After excluding 30% of the high-uncertainty samples, these metrics improve to an F1 score of 74.4% and an AUC of 85.6%.

CONCLUSION

Redirecting samples with high uncertainty to experts for subsequent diagnosis significantly decreases the rates of misdiagnosis, which highlights that uncertainty estimation is vital to ensure safe decision making for computer-aided early oral cancer diagnosis.

Trustworthy clinical AI solutions: A unified review of uncertainty quantification in Deep Learning models for medical image analysis

The full acceptance of Deep Learning (DL) models in the clinical field is rather low with respect to the quantity of high-performing solutions reported in the literature. End users are particularly reluctant to rely on the opaque predictions of DL models. Uncertainty quantification methods have been proposed in the literature as a potential solution, to reduce the black-box effect of DL models and increase the interpretability and the acceptability of the result by the final user. In this review, we propose an overview of the existing methods to quantify uncertainty associated with DL predictions. We focus on applications to medical image analysis, which present specific challenges due to the high dimensionality of images and their variable quality, as well as constraints associated with real-world clinical routine. Moreover, we discuss the concept of structural uncertainty, a corpus of methods to facilitate the alignment of segmentation uncertainty estimates with clinical attention. We then discuss the evaluation protocols to validate the relevance of uncertainty estimates. Finally, we highlight the open challenges for uncertainty quantification in the medical field.

Slideflow: deep learning for digital histopathology with real-time whole-slide visualization

Deep learning methods have emerged as powerful tools for analyzing histopathological images, but current methods are often specialized for specific domains and software environments, and few open-source options exist for deploying models in an interactive interface. Experimenting with different deep learning approaches typically requires switching software libraries and reprocessing data, reducing the feasibility and practicality of experimenting with new architectures. We developed a flexible deep learning library for histopathology called Slideflow, a package which supports a broad array of deep learning methods for digital pathology and includes a fast whole-slide interface for deploying trained models. Slideflow includes unique tools for whole-slide image data processing, efficient stain normalization and augmentation, weakly-supervised whole-slide classification, uncertainty quantification, feature generation, feature space analysis, and explainability. Whole-slide image processing is highly optimized, enabling whole-slide tile extraction at 40x magnification in 2.5 s per slide. The framework-agnostic data processing pipeline enables rapid experimentation with new methods built with either Tensorflow or PyTorch, and the graphical user interface supports real-time visualization of slides, predictions, heatmaps, and feature space characteristics on a variety of hardware devices, including ARM-based devices such as the Raspberry Pi.

A framework for evaluating clinical artificial intelligence systems without ground-truth annotations

A clinical artificial intelligence (AI) system is often validated on data withheld during its development. This provides an estimate of its performance upon future deployment on data in the wild; those currently unseen but are expected to be encountered in a clinical setting. However, estimating performance on data in the wild is complicated by distribution shift between data in the wild and withheld data and the absence of ground-truth annotations. Here, we introduce SUDO, a framework for evaluating AI systems on data in the wild. Through experiments on AI systems developed for dermatology images, histopathology patches, and clinical notes, we show that SUDO can identify unreliable predictions, inform the selection of models, and allow for the previously out-of-reach assessment of algorithmic bias for data in the wild without ground-truth annotations. These capabilities can contribute to the deployment of trustworthy and ethical AI systems in medicine.

Uncertainty-aware deep-learning model for prediction of supratentorial hematoma expansion from admission non-contrast head computed tomography scan

Hematoma expansion (HE) is a modifiable risk factor and a potential treatment target in patients with intracerebral hemorrhage (ICH). We aimed to train and validate deep-learning models for high-confidence prediction of supratentorial ICH expansion, based on admission non-contrast head Computed Tomography (CT). Applying Monte Carlo dropout and entropy of deep-learning model predictions, we estimated the model uncertainty and identified patients at high risk of HE with high confidence. Using the receiver operating characteristics area under the curve (AUC), we compared the deep-learning model prediction performance with multivariable models based on visual markers of HE determined by expert reviewers. We randomly split a multicentric dataset of patients (4-to-1) into training/cross-validation (n = 634) versus test (n = 159) cohorts. We trained and tested separate models for prediction of ≥6 mL and ≥3 mL ICH expansion. The deep-learning models achieved an AUC = 0.81 for high-confidence prediction of HE≥6 mL and AUC = 0.80 for prediction of HE≥3 mL, which were higher than visual maker models AUC = 0.69 for HE≥6 mL (p = 0.036) and AUC = 0.68 for HE≥3 mL (p = 0.043). Our results show that fully automated deep-learning models can identify patients at risk of supratentorial ICH expansion based on admission non-contrast head CT, with high confidence, and more accurately than benchmark visual markers.

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.

Benefits of spatial uncertainty aggregation for segmentation in digital pathology

Purpose

Uncertainty estimation has gained significant attention in recent years for its potential to enhance the performance of deep learning (DL) algorithms in medical applications and even potentially address domain shift challenges. However, it is not straightforward to incorporate uncertainty estimation with a DL system to achieve a tangible positive effect. The objective of our work is to evaluate if the proposed spatial uncertainty aggregation (SUA) framework may improve the effectiveness of uncertainty estimation in segmentation tasks. We evaluate if SUA boosts the observed correlation between the uncertainty estimates and false negative (FN) predictions. We also investigate if the observed benefits can translate to tangible improvements in segmentation performance.

Approach

Our SUA framework processes negative prediction regions from a segmentation algorithm and detects FNs based on an aggregated uncertainty score. It can be utilized with many existing uncertainty estimation methods to boost their performance. We compare the SUA framework with a baseline of processing individual pixel's uncertainty independently.

Results

The results demonstrate that SUA is able to detect FN regions. It achieved F β = 0.5 of 0.92 on the in-domain and 0.85 on the domain-shift test data compared with 0.81 and 0.48 achieved by the baseline uncertainty, respectively. We also demonstrate that SUA yields improved general segmentation performance compared with utilizing the baseline uncertainty.

Conclusions

We propose the SUA framework for incorporating and utilizing uncertainty estimates for FN detection in DL segmentation algorithms for histopathology. The evaluation confirms the benefits of our approach compared with assessing pixel uncertainty independently.

A Graph Neural Network Model with a Transparent Decision-Making Process Defines the Applicability Domain for Environmental Estrogen Screening

The application of deep learning (DL) models for screening environmental estrogens (EEs) for the sound management of chemicals has garnered significant attention. However, the currently available DL model for screening EEs lacks both a transparent decision-making process and effective applicability domain (AD) characterization, making the reliability of its prediction results uncertain and limiting its practical applications. To address this issue, a graph neural network (GNN) model was developed to screen EEs, achieving accuracy rates of 88.9% and 92.5% on the internal and external test sets, respectively. The decision-making process of the GNN model was explored through the network-like similarity graphs (NSGs) based on the model features (FT). We discovered that the accuracy of the predictions is dependent on the feature distribution of compounds in NSGs. An AD characterization method called ADFT was proposed, which excludes predictions falling outside of the model's prediction range, leading to a 15% improvement in the F1 score of the GNN model. The GNN model with the AD method may serve as an efficient tool for screening EEs, identifying 800 potential EEs in the Inventory of Existing Chemical Substances of China. Additionally, this study offers new insights into comprehending the decision-making process of DL models.

Uncertainty-inspired open set learning for retinal anomaly identification

Failure to recognize samples from the classes unseen during training is a major limitation of artificial intelligence in the real-world implementation for recognition and classification of retinal anomalies. We establish an uncertainty-inspired open set (UIOS) model, which is trained with fundus images of 9 retinal conditions. Besides assessing the probability of each category, UIOS also calculates an uncertainty score to express its confidence. Our UIOS model with thresholding strategy achieves an F1 score of 99.55%, 97.01% and 91.91% for the internal testing set, external target categories (TC)-JSIEC dataset and TC-unseen testing set, respectively, compared to the F1 score of 92.20%, 80.69% and 64.74% by the standard AI model. Furthermore, UIOS correctly predicts high uncertainty scores, which would prompt the need for a manual check in the datasets of non-target categories retinal diseases, low-quality fundus images, and non-fundus images. UIOS provides a robust method for real-world screening of retinal anomalies.

Application of uncertainty quantification to artificial intelligence in healthcare: A review of last decade (2013-2023)

Uncertainty estimation in healthcare involves quantifying and understanding the inherent uncertainty or variability associated with medical predictions, diagnoses, and treatment outcomes. In this era of Artificial Intelligence (AI) models, uncertainty estimation becomes vital to ensure safe decision-making in the medical field. Therefore, this review focuses on the application of uncertainty techniques to machine and deep learning models in healthcare. A systematic literature review was conducted using the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines. Our analysis revealed that Bayesian methods were the predominant technique for uncertainty quantification in machine learning models, with Fuzzy systems being the second most used approach. Regarding deep learning models, Bayesian methods emerged as the most prevalent approach, finding application in nearly all aspects of medical imaging. Most of the studies reported in this paper focused on medical images, highlighting the prevalent application of uncertainty quantification techniques using deep learning models compared to machine learning models. Interestingly, we observed a scarcity of studies applying uncertainty quantification to physiological signals. Thus, future research on uncertainty quantification should prioritize investigating the application of these techniques to physiological signals. Overall, our review highlights the significance of integrating uncertainty techniques in healthcare applications of machine learning and deep learning models. This can provide valuable insights and practical solutions to manage uncertainty in real-world medical data, ultimately improving the accuracy and reliability of medical diagnoses and treatment recommendations.

Development of an artificial intelligence bacteremia prediction model and evaluation of its impact on physician predictions focusing on uncertainty

Prediction of bacteremia is a clinically important but challenging task. An artificial intelligence (AI) model has the potential to facilitate early bacteremia prediction, aiding emergency department (ED) physicians in making timely decisions and reducing unnecessary medical costs. In this study, we developed and externally validated a Bayesian neural network-based AI bacteremia prediction model (AI-BPM). We also evaluated its impact on physician predictive performance considering both AI and physician uncertainties using historical patient data. A retrospective cohort of 15,362 adult patients with blood cultures performed in the ED was used to develop the AI-BPM. The AI-BPM used structured and unstructured text data acquired during the early stage of ED visit, and provided both the point estimate and 95% confidence interval (CI) of its predictions. High AI-BPM uncertainty was defined as when the predetermined bacteremia risk threshold (5%) was included in the 95% CI of the AI-BPM prediction, and low AI-BPM uncertainty was when it was not included. In the temporal validation dataset (N = 8,188), the AI-BPM achieved area under the receiver operating characteristic curve (AUC) of 0.754 (95% CI 0.737-0.771), sensitivity of 0.917 (95% CI 0.897-0.934), and specificity of 0.340 (95% CI 0.330-0.351). In the external validation dataset (N = 7,029), the AI-BPM's AUC was 0.738 (95% CI 0.722-0.755), sensitivity was 0.927 (95% CI 0.909-0.942), and specificity was 0.319 (95% CI 0.307-0.330). The AUC of the post-AI physicians predictions (0.703, 95% CI 0.654-0.753) was significantly improved compared with that of the pre-AI predictions (0.639, 95% CI 0.585-0.693; p-value < 0.001) in the sampled dataset (N = 1,000). The AI-BPM especially improved the predictive performance of physicians in cases with high physician uncertainty (low subjective confidence) and low AI-BPM uncertainty. Our results suggest that the uncertainty of both the AI model and physicians should be considered for successful AI model implementation.

Deep Learning for Lung Cancer Diagnosis, Prognosis and Prediction Using Histological and Cytological Images: A Systematic Review

Lung cancer is one of the deadliest cancers worldwide, with a high incidence rate, especially in tobacco smokers. Lung cancer accurate diagnosis is based on distinct histological patterns combined with molecular data for personalized treatment. Precise lung cancer classification from a single H&E slide can be challenging for a pathologist, requiring most of the time additional histochemical and special immunohistochemical stains for the final pathology report. According to WHO, small biopsy and cytology specimens are the available materials for about 70% of lung cancer patients with advanced-stage unresectable disease. Thus, the limited available diagnostic material necessitates its optimal management and processing for the completion of diagnosis and predictive testing according to the published guidelines. During the new era of Digital Pathology, Deep Learning offers the potential for lung cancer interpretation to assist pathologists' routine practice. Herein, we systematically review the current Artificial Intelligence-based approaches using histological and cytological images of lung cancer. Most of the published literature centered on the distinction between lung adenocarcinoma, lung squamous cell carcinoma, and small cell lung carcinoma, reflecting the realistic pathologist's routine. Furthermore, several studies developed algorithms for lung adenocarcinoma predominant architectural pattern determination, prognosis prediction, mutational status characterization, and PD-L1 expression status estimation.

Unleashing the potential of AI for pathology: challenges and recommendations

Computational pathology is currently witnessing a surge in the development of AI techniques, offering promise for achieving breakthroughs and significantly impacting the practices of pathology and oncology. These AI methods bring with them the potential to revolutionize diagnostic pipelines as well as treatment planning and overall patient care. Numerous peer-reviewed studies reporting remarkable performance across diverse tasks serve as a testimony to the potential of AI in the field. However, widespread adoption of these methods in clinical and pre-clinical settings still remains a challenge. In this review article, we present a detailed analysis of the major obstacles encountered during the development of effective models and their deployment in practice. We aim to provide readers with an overview of the latest developments, assist them with insights into identifying some specific challenges that may require resolution, and suggest recommendations and potential future research directions. © 2023 The Authors. The Journal of Pathology published by John Wiley & Sons Ltd on behalf of The Pathological Society of Great Britain and Ireland.

Proceedings of the North American Society of Head and Neck Pathology Companion Meeting, New Orleans, LA, March 12, 2023: Oral Cavity Dysplasia: Why Does Histologic Grading Continue to be Contentious?

Head and neck squamous cell carcinoma (HNSCC) is the world's 6th most common malignancy. Oral cavity SCC (OCSCC) represents approximately one third of the HNSCC cases diagnosed annually in the United States. Despite therapeutic advances, OCSCC is frequently lethal, with a modest 5-year survival. Because OCSCC is often preceded by premalignant lesions, it is an ideal disease for screening initiatives. The conventional visual and tactile exam (CVTE), coupled with a tissue biopsy, remains the gold standard. However, CVTE alone cannot reliably differentiate between reactive/inflammatory and dysplastic lesions. Further, the histologic diagnosis of dysplasia is subjective in nature and a highly imperfect predictor of malignant transformation. This prognostic uncertainty creates a significant clinical management dilemma-watchful waiting with increased patient psychological and economic burdens versus unnecessary aggressive treatment. As such, the development and validation of novel diagnostic platforms such as Artificial Intelligence (AI) and prognostic molecular biomarkers may help address these critical unmet clinical needs.

Application of simultaneous uncertainty quantification for image segmentation with probabilistic deep learning: Performance benchmarking of oropharyngeal cancer target delineation as a use-case

Background

Oropharyngeal cancer (OPC) is a widespread disease, with radiotherapy being a core treatment modality. Manual segmentation of the primary gross tumor volume (GTVp) is currently employed for OPC radiotherapy planning, but is subject to significant interobserver variability. Deep learning (DL) approaches have shown promise in automating GTVp segmentation, but comparative (auto)confidence metrics of these models predictions has not been well-explored. Quantifying instance-specific DL model uncertainty is crucial to improving clinician trust and facilitating broad clinical implementation. Therefore, in this study, probabilistic DL models for GTVp auto-segmentation were developed using large-scale PET/CT datasets, and various uncertainty auto-estimation methods were systematically investigated and benchmarked.

Methods

We utilized the publicly available 2021 HECKTOR Challenge training dataset with 224 co-registered PET/CT scans of OPC patients with corresponding GTVp segmentations as a development set. A separate set of 67 co-registered PET/CT scans of OPC patients with corresponding GTVp segmentations was used for external validation. Two approximate Bayesian deep learning methods, the MC Dropout Ensemble and Deep Ensemble, both with five submodels, were evaluated for GTVp segmentation and uncertainty performance. The segmentation performance was evaluated using the volumetric Dice similarity coefficient (DSC), mean surface distance (MSD), and Hausdorff distance at 95% (95HD). The uncertainty was evaluated using four measures from literature: coefficient of variation (CV), structure expected entropy, structure predictive entropy, and structure mutual information, and additionally with our novel Dice-risk measure. The utility of uncertainty information was evaluated with the accuracy of uncertainty-based segmentation performance prediction using the Accuracy vs Uncertainty (AvU) metric, and by examining the linear correlation between uncertainty estimates and DSC. In addition, batch-based and instance-based referral processes were examined, where the patients with high uncertainty were rejected from the set. In the batch referral process, the area under the referral curve with DSC (R-DSC AUC) was used for evaluation, whereas in the instance referral process, the DSC at various uncertainty thresholds were examined.

Results

Both models behaved similarly in terms of the segmentation performance and uncertainty estimation. Specifically, the MC Dropout Ensemble had 0.776 DSC, 1.703 mm MSD, and 5.385 mm 95HD. The Deep Ensemble had 0.767 DSC, 1.717 mm MSD, and 5.477 mm 95HD. The uncertainty measure with the highest DSC correlation was structure predictive entropy with correlation coefficients of 0.699 and 0.692 for the MC Dropout Ensemble and the Deep Ensemble, respectively. The highest AvU value was 0.866 for both models. The best performing uncertainty measure for both models was the CV which had R-DSC AUC of 0.783 and 0.782 for the MC Dropout Ensemble and Deep Ensemble, respectively. With referring patients based on uncertainty thresholds from 0.85 validation DSC for all uncertainty measures, on average the DSC improved from the full dataset by 4.7% and 5.0% while referring 21.8% and 22% patients for MC Dropout Ensemble and Deep Ensemble, respectively.

Conclusion

We found that many of the investigated methods provide overall similar but distinct utility in terms of predicting segmentation quality and referral performance. These findings are a critical first-step towards more widespread implementation of uncertainty quantification in OPC GTVp segmentation.