Perspectives on label-free microscopy of heterogeneous and dynamic biological systems

Significance

Advancements in label-free microscopy could provide real-time, non-invasive imaging with unique sources of contrast and automated standardized analysis to characterize heterogeneous and dynamic biological processes. These tools would overcome challenges with widely used methods that are destructive (e.g., histology, flow cytometry) or lack cellular resolution (e.g., plate-based assays, whole animal bioluminescence imaging).

Aim

This perspective aims to (1) justify the need for label-free microscopy to track heterogeneous cellular functions over time and space within unperturbed systems and (2) recommend improvements regarding instrumentation, image analysis, and image interpretation to address these needs.

Approach

Three key research areas (cancer research, autoimmune disease, and tissue and cell engineering) are considered to support the need for label-free microscopy to characterize heterogeneity and dynamics within biological systems. Based on the strengths (e.g., multiple sources of molecular contrast, non-invasive monitoring) and weaknesses (e.g., imaging depth, image interpretation) of several label-free microscopy modalities, improvements for future imaging systems are recommended.

Conclusion

Improvements in instrumentation including strategies that increase resolution and imaging speed, standardization and centralization of image analysis tools, and robust data validation and interpretation will expand the applications of label-free microscopy to study heterogeneous and dynamic biological systems.

Challenge, integration, and change: ChatGPT and future anatomical education

With the vigorous development of ChatGPT and its application in the field of education, a new era of the collaborative development of human and artificial intelligence and the symbiosis of education has come. Integrating artificial intelligence (AI) into medical education has the potential to revolutionize it. Large language models, such as ChatGPT, can be used as virtual teaching aids to provide students with individualized and immediate medical knowledge, and conduct interactive simulation learning and detection. In this paper, we discuss the application of ChatGPT in anatomy teaching and its various application levels based on our own teaching experiences, and discuss the advantages and disadvantages of ChatGPT in anatomy teaching. ChatGPT increases student engagement and strengthens students' ability to learn independently. At the same time, ChatGPT faces many challenges and limitations in medical education. Medical educators must keep pace with the rapid changes in technology, taking into account ChatGPT's impact on curriculum design, assessment strategies and teaching methods. Discussing the application of ChatGPT in medical education, especially anatomy teaching, is helpful to the effective integration and application of artificial intelligence tools in medical education.

The use of artificial intelligence to assess diabetic eye disease among the Greenlandic population

Background: Retina fundus images conducted in Greenland are telemedically assessed for diabetic retinopathy by ophthalmological nurses in Denmark. Applying an AI grading solution, in a Greenlandic setting, could potentially improve the efficiency and cost-effectiveness of DR screening.Method: We developed an AI model using retina fundus photos, performed on persons registered with diabetes in Greenland and Denmark, using Optos® ultra wide-field scanning laser ophthalmoscope, graded according to ICDR.Using the ResNet50 network we compared the model's ability to distinguish between different images of ICDR severity levels in a confusion matrix.Results: Comparing images with ICDR level 0 to images of ICDR level 4 resulted in an accuracy of 0.9655, AUC of 0.9905, sensitivity and specificity of 96.6%.Comparing ICDR levels 0,1,2 with ICDR levels 3,4, we achieved a performance with an accuracy of 0.8077, an AUC of 0.8728, a sensitivity of 84.6% and a specificity of 78.8%. For the other comparisons, we achieved a modest performance.Conclusion: We developed an AI model using Greenlandic data, to automatically detect DR on Optos retina fundus images. The sensitivity and specificity were too low for our model to be applied directly in a clinical setting, thus optimising the model should be prioritised.

Text mining of hypertension researches in the west Asia region: a 12-year trend analysis

More than half of the world population lives in Asia and hypertension (HTN) is the most prevalent risk factor found in Asia. There are numerous articles published about HTN in Eastern Mediterranean Region (EMRO) and artificial intelligence (AI) methods can analyze articles and extract top trends in each country. Present analysis uses Latent Dirichlet allocation (LDA) as an algorithm of topic modeling (TM) in text mining, to obtain subjective topic-word distribution from the 2790 studies over the EMRO. The period of checked studied is last 12 years and results of LDA analyses show that HTN researches published in EMRO discuss on changes in BP and the factors affecting it. Among the countries in the region, most of these articles are related to I.R Iran and Egypt, which have an increasing trend from 2017 to 2018 and reached the highest level in 2021. Meanwhile, Iraq and Lebanon have been conducting research since 2010. The EMRO word cloud illustrates 'BMI', 'mortality', 'age', and 'meal', which represent important indicators, dangerous outcomes of high BP, and gender of HTN patients in EMRO, respectively.

Advances in Activity/Property Prediction from Chemical Structures

Recent technological advancement in AI modeling of molecular property databases has significantly expanded the opportunities for drug design and development. Quantitative structure-activity relationships (QSARs) are shown to provide more accurate predictions with regards to biological activity as well as toxicological assessment. By using a combination of in-silico models or by combining disparate structure-activity databases, researchers have been able to improve accuracy for a variety of drug discovery and analysis methods, generating viable compounds, which in certain cases, can be synthesized and further studied in vitro to find candidates for potential development. Additionally, the development of compounds of determined toxicology can be discontinued earlier, allowing alternative routes to be evaluated, preventing wasted time and resources. Although the progress that has been made is tremendous, expert review is still necessary for most in-silico generated predictions. Regardless, the scientific community continues to move ever closer to completely automated drug discovery and evaluation.

The new era of artificial intelligence in neuroradiology: current research and promising tools

Radiology has a number of characteristics that make it an especially suitable medical discipline for early artificial intelligence (AI) adoption. These include having a well-established digital workflow, standardized protocols for image storage, and numerous well-defined interpretive activities. The more than 200 commercial radiologic AI-based products recently approved by the Food and Drug Administration (FDA) to assist radiologists in a number of narrow image-analysis tasks such as image enhancement, workflow triage, and quantification, corroborate this observation. However, in order to leverage AI to boost efficacy and efficiency, and to overcome substantial obstacles to widespread successful clinical use of these products, radiologists should become familiarized with the emerging applications in their particular areas of expertise. In light of this, in this article we survey the existing literature on the application of AI-based techniques in neuroradiology, focusing on conditions such as vascular diseases, epilepsy, and demyelinating and neurodegenerative conditions. We also introduce some of the algorithms behind the applications, briefly discuss a few of the challenges of generalization in the use of AI models in neuroradiology, and skate over the most relevant commercially available solutions adopted in clinical practice. If well designed, AI algorithms have the potential to radically improve radiology, strengthening image analysis, enhancing the value of quantitative imaging techniques, and mitigating diagnostic errors.

Binary classification model of machine learning detected altered gut integrity in controlled-cortical impact model of traumatic brain injury

Aim of the study: To examine the effect of controlled-cortical impact (CCI), a preclinical model of traumatic brain injury (TBI), on intestinal integrity using a binary classification model of machine learning (ML).Materials and methods: Adult, male C57BL/6 mice were subjected to CCI surgery using a stereotaxic impactor (Impact One™). The rotarod and hot-plate tests were performed to assess the neurological deficits.Results: Mice underwent CCI displayed a remarkable neurological deficit as noticed by decreased latency to fall and lesser paw withdrawal latency in rotarod and hot plate test, respectively. Animals were sacrificed 3 days post-injury (dpi). The colon sections were stained with hematoxylin and eosin (H&E) to integrate with machinery tool-based algorithms. Several stained colon images were captured to build a dataset for ML model to predict the impact of CCI vs sham procedure. The best results were obtained with VGG16 features with SVM RBF kernel and VGG16 features with stacked fully connected layers on top. We achieved a test accuracy of 84% and predicted the disrupted gut permeability and epithelium wall of colon in CCI group as compared to sham-operated mice.Conclusion: We suggest that ML may become an important tool in the development of preclinical TBI model and discovery of newer therapeutics.

Using artificial intelligence platforms to support student learning in physiology

The advent of artificial intelligence (AI)-powered language models presents new opportunities and challenges in education. By teaching students how to craft prompts that elicit insightful responses, faculty can scaffold activities where AI acts as a supplemental resource to amplify critical thinking and support student learning. Ongoing dialogue and iteration focused on ethical usage norms can achieve the right balance between emerging technology and foundational skills development. With care and intention, AI-assisted study tactics offer students personalized support while adhering to academic standards. While AI-powered tools provide many positive opportunities, students and faculty need to learn about and use them responsibly and ethically, not as replacements for required thinking and effort. Before implementing these AI tools for studying biology, there are several key things to discuss with students. This article outlines several ways that students can employ these tools to support better learning along with a set of guidelines for all to be wary of when implementing these in an academic setting.NEW & NOTEWORTHY Utilizing of artificial intelligence tools offers a promising new technology to support student learning. This article outlines several ways that students can employ these tools to support better learning along with a set of guidelines for all to be wary of when implementing these in an academic setting.

Schistocyte detection in artificial intelligence age

Schistocytes are fragmented red blood cells produced as a result of mechanical damage to erythrocytes, usually due to microangiopathic thrombotic diseases or mechanical factors. The early laboratory detection of schistocytes has a critical impact on the timely diagnosis, effective treatment, and positive prognosis of diseases such as thrombocytopenic purpura and hemolytic uremic syndrome. Due to the rapid development of science and technology, laboratory hematology has also advanced. The accuracy and efficiency of tests performed by fully automated hematology analyzers and fully automated morphology analyzers have been considerably improved. In recent years, substantial improvements in computing power and machine learning (ML) algorithm development have dramatically extended the limits of the potential of autonomous machines. The rapid development of machine learning and artificial intelligence (AI) has led to the iteration and upgrade of automated detection of schistocytes. However, along with significantly facilitated operation processes, AI has brought challenges. This review summarizes the progress in laboratory schistocyte detection, the relationship between schistocytes and clinical diseases, and the progress of AI in the detection of schistocytes. In addition, current challenges and possible solutions are discussed, as well as the great potential of AI techniques for schistocyte testing in peripheral blood.

Multicenter prospective blinded melanoma detection study with a handheld elastic scattering spectroscopy device

Background

The elastic scattering spectroscopy (ESS) device (DermaSensor Inc., Miami, FL) is a noninvasive, painless, adjunctive tool for skin cancer detection.

Objectives

To investigate the performance of the ESS device in the detection of melanoma.

Methods

A prospective, investigator-blinded, multicenter study was conducted at 8 United States (US) and 2 Australian sites. All eligible skin lesions were clinically concerning for melanoma, examined with the ESS device, subsequently biopsied according to dermatologists' standard of care, and evaluated with histopathology. A total of 311 participants with 440 lesions were enrolled, including 44 melanomas (63.6% in situ and 36.4% invasive) and 44 severely dysplastic nevi.

Results

The observed sensitivity of the ESS device for melanoma detection was 95.5% (95% CI, 84.5% to 98.8%, 42 of 44 melanomas), and the observed specificity was 32.5% (95% CI, 27.2% to 38.3%). The positive and negative predictive values were 16.0% and 98.1%, respectively.

Limitations

The device was tested in a high-risk population with lesions selected for biopsy based on clinical and dermoscopic assessments of board-certified dermatologists. Most enrolled lesions were pigmented.

Conclusion

The ESS device's high sensitivity and NPV for the detection of melanoma suggest the device may be a useful adjunctive, point-of-care tool for melanoma detection.

Artificial Intelligence for Anesthesiology Board-Style Examination Questions: Role of Large Language Models

New artificial intelligence tools have been developed that have implications for medical usage. Large language models (LLMs), such as the widely used ChatGPT developed by OpenAI, have not been explored in the context of anesthesiology education. Understanding the reliability of various publicly available LLMs for medical specialties could offer insight into their understanding of the physiology, pharmacology, and practical applications of anesthesiology. An exploratory prospective review was conducted using 3 commercially available LLMs--OpenAI's ChatGPT GPT-3.5 version (GPT-3.5), OpenAI's ChatGPT GPT-4 (GPT-4), and Google's Bard--on questions from a widely used anesthesia board examination review book. Of the 884 eligible questions, the overall correct answer rates were 47.9% for GPT-3.5, 69.4% for GPT-4, and 45.2% for Bard. GPT-4 exhibited significantly higher performance than both GPT-3.5 and Bard (p = 0.001 and p < 0.001, respectively). None of the LLMs met the criteria required to secure American Board of Anesthesiology certification, according to the 70% passing score approximation. GPT-4 significantly outperformed GPT-3.5 and Bard in terms of overall performance, but lacked consistency in providing explanations that aligned with scientific and medical consensus. Although GPT-4 shows promise, current LLMs are not sufficiently advanced to answer anesthesiology board examination questions with passing success. Further iterations and domain-specific training may enhance their utility in medical education.

Enhancing thalassemia gene carrier identification in non-anemic populations using artificial intelligence erythrocyte morphology analysis and machine learning

BACKGROUND

Non-anemic thalassemia trait (TT) accounted for a high proportion of TT cases in South China.

OBJECTIVE

To use artificial intelligence (AI) analysis of erythrocyte morphology and machine learning (ML) to identify TT gene carriers in a non-anemic population.

METHODS

Digital morphological data from 76 TT gene carriers and 97 controls were collected. The AI technology-based Mindray MC-100i was used to quantitatively analyze the percentage of abnormal erythrocytes. Further, ML was used to construct a prediction model.

RESULTS

Non-anemic TT carriers accounted for over 60% of the TT cases. Random Forest was selected as the prediction model and named TT@Normal. The TT@Normal algorithm showed outstanding performance in the training, validation, and external validation sets and could efficiently identify TT carriers in the non-anemic population. The top three weights in the TT@Normal model were the target cells, microcytes, and teardrop cells. Elevated percentages of abnormal erythrocytes should raise a strong suspicion of being a TT gene carrier. TT@Normal could be promoted and used as a visualization and sharing tool. It is accessible through a URL link and can be used by medical staff online to predict the possibility of TT gene carriage in a non-anemic population.

CONCLUSIONS

The ML-based model TT@Normal could efficiently identify TT carriers in non-anemic people. Elevated percentages of target cells, microcytes, and teardrop cells should raise a strong suspicion of being a TT gene carrier.

PRERISK: A Personalized, Artificial Intelligence-Based and Statistically-Based Stroke Recurrence Predictor for Recurrent Stroke

BACKGROUND

Predicting stroke recurrence for individual patients is difficult, but individualized prediction may improve stroke survivors' engagement in self-care. We developed PRERISK: a statistical and machine learning classifier to predict individual risk of stroke recurrence.

METHODS

We analyzed clinical and socioeconomic data from a prospectively collected public health care-based data set of 41 975 patients admitted with stroke diagnosis in 88 public health centers over 6 years (2014-2020) in Catalonia-Spain. A new stroke diagnosis at least 24 hours after the index event was considered as a recurrent stroke, which was considered as our outcome of interest. We trained several supervised machine learning models to provide individualized risk over time and compared them with a Cox regression model. Models were trained to predict early, late, and long-term recurrence risk, within 90, 91 to 365, and >365 days, respectively. C statistics and area under the receiver operating characteristic curve were used to assess the accuracy of the models.

RESULTS

Overall, 16.21% (5932 of 36 114) of patients had stroke recurrence during a median follow-up of 2.69 years. The most powerful predictors of stroke recurrence were time from previous stroke, Barthel Index, atrial fibrillation, dyslipidemia, age, diabetes, and sex, which were used to create a simplified model with similar performance, together with modifiable vascular risk factors (glycemia, body mass index, high blood pressure, cholesterol, tobacco dependence, and alcohol abuse). The areas under the receiver operating characteristic curve were 0.76 (95% CI, 0.74-0.77), 0.60 (95% CI, 0.58-0.61), and 0.71 (95% CI, 0.69-0.72) for early, late, and long-term recurrence risk, respectively. The areas under the receiver operating characteristic curve of the Cox risk class probability were 0.73 (95% CI, 0.72-0.75), 0.59 (95% CI, 0.57-0.61), and 0.67 (95% CI, 0.66-0.70); machine learning approaches (random forest and AdaBoost) showed statistically significant improvement (P<0.05) over the Cox model for the 3 recurrence time periods. Stroke recurrence curves can be simulated for each patient under different degrees of control of modifiable factors.

CONCLUSIONS

PRERISK is a novel approach that provides a personalized and fairly accurate risk prediction of stroke recurrence over time. The model has the potential to incorporate dynamic control of risk factors.

Applications of artificial intelligence-enabled robots and chatbots in ophthalmology: recent advances and future trends

PURPOSE OF REVIEW

Recent advances in artificial intelligence (AI), robotics, and chatbots have brought these technologies to the forefront of medicine, particularly ophthalmology. These technologies have been applied in diagnosis, prognosis, surgical operations, and patient-specific care in ophthalmology. It is thus both timely and pertinent to assess the existing landscape, recent advances, and trajectory of trends of AI, AI-enabled robots, and chatbots in ophthalmology.

RECENT FINDINGS

Some recent developments have integrated AI enabled robotics with diagnosis, and surgical procedures in ophthalmology. More recently, large language models (LLMs) like ChatGPT have shown promise in augmenting research capabilities and diagnosing ophthalmic diseases. These developments may portend a new era of doctor-patient-machine collaboration.

SUMMARY

Ophthalmology is undergoing a revolutionary change in research, clinical practice, and surgical interventions. Ophthalmic AI-enabled robotics and chatbot technologies based on LLMs are converging to create a new era of digital ophthalmology. Collectively, these developments portend a future in which conventional ophthalmic knowledge will be seamlessly integrated with AI to improve the patient experience and enhance therapeutic outcomes.

Development of Artificial Intelligence Image Classification Models for Determination of Umbilical Cord Vascular Anomalies

OBJECTIVE

The goal of this work was to develop robust techniques for the processing and identification of SUA using artificial intelligence (AI) image classification models.

METHODS

Ultrasound images obtained retrospectively were analyzed for blinding, text removal, AI training, and image prediction. After developing and testing text removal methods, a small n-size study (40 images) using fastai/PyTorch to classify umbilical cord images. This data set was expanded to 286 lateral-CFI images that were used to compare: different neural network performance, diagnostic value, and model predictions.

RESULTS

AI-Optical Character Recognition method was superior in its ability to remove text from images. The small n-size mixed single umbilical artery determination data set was tested with a pretrained ResNet34 neural network and obtained and error rate average of 0.083 (n = 3). The expanded data set was then tested with several AI models. The majority of the tested networks were able to obtain an average error rate of <0.15 with minimal modifications. The ResNet34-default performed the best with: an image-classification error rate of 0.0175, sensitivity of 1.00, specificity of 0.97, and ability to correctly infer classification.

CONCLUSION

This work provides a robust framework for ultrasound image AI classifications. AI could successfully classify umbilical cord types of ultrasound image study with excellent diagnostic value. Together this study provides a reproducible framework to develop AI-specific ultrasound classification of umbilical cord or other diagnoses to be used in conjunction with physicians for optimal patient care.

Utilizing Deep Learning and Computed Tomography to Determine Pulmonary Nodule Activity in Patients With Nontuberculous Mycobacterial-Lung Disease

PURPOSE

To develop and evaluate a deep convolutional neural network (DCNN) model for the classification of acute and chronic lung nodules from nontuberculous mycobacterial-lung disease (NTM-LD) on computed tomography (CT).

MATERIALS AND METHODS

We collected a data set of 650 nodules (316 acute and 334 chronic) from the CT scans of 110 patients with NTM-LD. The data set was divided into training, validation, and test sets in a ratio of 4:1:1. Bounding boxes were used to crop the 2D CT images down to the area of interest. A DCNN model was built using 11 convolutional layers and trained on these images. The performance of the model was evaluated on the hold-out test set and compared with that of 3 radiologists who independently reviewed the images.

RESULTS

The DCNN model achieved an area under the receiver operating characteristic curve of 0.806 for differentiating acute and chronic NTM-LD nodules, corresponding to sensitivity, specificity, and accuracy of 76%, 68%, and 72%, respectively. The performance of the model was comparable to that of the 3 radiologists, who had area under the receiver operating characteristic curve, sensitivity, specificity, and accuracy of 0.693 to 0.771, 61% to 82%, 59% to 73%, and 60% to 73%, respectively.

CONCLUSIONS

This study demonstrated the feasibility of using a DCNN model for the classification of the activity of NTM-LD nodules on chest CT. The model performance was comparable to that of radiologists. This approach can potentially and efficiently improve the diagnosis and management of NTM-LD.

Leading in the development, standardised evaluation, and adoption of artificial intelligence in clinical practice: regional anaesthesia as an example

A recent study by Suissa and colleagues explored the clinical relevance of a medical image segmentation metric (Dice metric) commonly used in the field of artificial intelligence (AI). They showed that pixel-wise agreement for physician identification of structures on ultrasound images is variable, and a relatively low Dice metric (0.34) correlated to a substantial agreement on subjective clinical assessment. We highlight the need to bring structure and clinical perspective to the evaluation of medical AI, which clinicians are best placed to direct.

Natural language processing of clinical notes enables early inborn error of immunity risk ascertainment

Background

There are now approximately 450 discrete inborn errors of immunity (IEI) described; however, diagnostic rates remain suboptimal. Use of structured health record data has proven useful for patient detection but may be augmented by natural language processing (NLP). Here we present a machine learning model that can distinguish patients from controls significantly in advance of ultimate diagnosis date.

Objective

We sought to create an NLP machine learning algorithm that could identify IEI patients early during the disease course and shorten the diagnostic odyssey.

Methods

Our approach involved extracting a large corpus of IEI patient clinical-note text from a major referral center's electronic health record (EHR) system and a matched control corpus for comparison. We built text classifiers with simple machine learning methods and trained them on progressively longer time epochs before date of diagnosis.

Results

The top performing NLP algorithm effectively distinguished cases from controls robustly 36 months before ultimate clinical diagnosis (area under precision recall curve > 0.95). Corpus analysis demonstrated that statistically enriched, IEI-relevant terms were evident 24+ months before diagnosis, validating that clinical notes can provide a signal for early prediction of IEI.

Conclusion

Mining EHR notes with NLP holds promise for improving early IEI patient detection.

Evaluation of Oropharyngeal Cancer Information from Revolutionary Artificial Intelligence Chatbot

OBJECTIVE

With burgeoning popularity of artificial intelligence-based chatbots, oropharyngeal cancer patients now have access to a novel source of medical information. Because chatbot information is not reviewed by experts, we sought to evaluate an artificial intelligence-based chatbot's oropharyngeal cancer-related information for accuracy.

METHODS

Fifteen oropharyngeal cancer-related questions were developed and input into ChatGPT version 3.5. Four physician-graders independently assessed accuracy, comprehensiveness, and similarity to a physician response using 5-point Likert scales. Responses graded lower than three were then critiqued by physician-graders. Critiques were analyzed using inductive thematic analysis. Readability of responses was assessed using Flesch Reading Ease (FRE) and Flesch-Kincaid Reading Grade Level (FKRGL) scales.

RESULTS

Average accuracy, comprehensiveness, and similarity to a physician response scores were 3.88 (SD = 0.99), 3.80 (SD = 1.14), and 3.67 (SD = 1.08), respectively. Posttreatment-related questions were most accurate, comprehensive, and similar to a physician response, followed by treatment-related, then diagnosis-related questions. Posttreatment-related questions scored significantly higher than diagnosis-related questions in all three domains (p < 0.01). Two themes of the physician critiques were identified: suboptimal education value and potential to misinform patients. The mean FRE and FKRGL scores both indicated greater than an 11th grade readability level-higher than the 6th grade level recommended for patients.

CONCLUSION

ChatGPT responses may not educate patients to an appropriate degree, could outright misinform them, and read at a more difficult grade level than is recommended for patient material. As oropharyngeal cancer patients represent a vulnerable population facing complex, life-altering diagnoses, and treatments, they should be cautious when consuming chatbot-generated medical information.

LEVEL OF EVIDENCE

NA Laryngoscope, 134:2252-2257, 2024.

Stenosis Detection and Quantification of Coronary Artery Using Machine Learning and Deep Learning

The aim of this review is to introduce some applications of artificial intelligence (AI) algorithms for the detection and quantification of coronary stenosis using computed tomography angiography (CTA). The realization of automatic/semi-automatic stenosis detection and quantification includes the following steps: vessel central axis extraction, vessel segmentation, stenosis detection, and quantification. Many new AI techniques, such as machine learning and deep learning, have been widely used in medical image segmentation and stenosis detection. This review also summarizes the recent progress regarding coronary stenosis detection and quantification, and discusses the development trends in this field. Through evaluation and comparison, researchers can better understand the research frontier in related fields, compare the advantages and disadvantages of various methods, and better optimize the new technologies. Machine learning and deep learning will promote the process of automatic detection and quantification of coronary artery stenosis. However, the machine learning and the deep learning methods need a large amount of data, so they also face some challenges because of the lack of professional image annotations (manually add labels by experts).

Acta Ophthalmologica 100 years-Overview of selected articles during Acta Ophthalmologica history

We selected and discussed 10 articles in Acta Ophthalmologica since 1923 that changed clinical ophthalmology and treatment protocols, or provided novel findings and perspectives. We are aware that the selection of articles may be debatable and we invite readers to suggest other significant Acta articles. For historians, the article archive of Acta Ophthalmologica is located in Copenhagen.

R WE ready for reimbursement? A round up of developments in real-world evidence relating to health technology assessment: part 15

In this latest update we discuss real-world evidence (RWE) guidance from the leading oncology professional societies, the American Society of Clinical Oncology and the European Society for Medical Oncology, and the PRINCIPLED practical guide on the design and analysis of causal RWE studies.

Machine learning based local recurrence prediction in colorectal cancer using polarized light imaging

Significance

Current treatment for stage III colorectal cancer (CRC) patients involves surgery that may not be sufficient in many cases, requiring additional adjuvant systemic therapy. Identification of this latter cohort that is likely to recur following surgery is key to better personalized therapy selection, but there is a lack of proper quantitative assessment tools for potential clinical adoption.

Aim

The purpose of this study is to employ Mueller matrix (MM) polarized light microscopy in combination with supervised machine learning (ML) to quantitatively analyze the prognostic value of peri-tumoral collagen in CRC in relation to 5-year local recurrence (LR).

Approach

A simple MM microscope setup was used to image surgical resection samples acquired from stage III CRC patients. Various potential biomarkers of LR were derived from MM elements via decomposition and transformation operations. These were used as features by different supervised ML models to distinguish samples from patients that locally recurred 5 years later from those that did not.

Results

Using the top five most prognostic polarimetric biomarkers ranked by their relevant feature importances, the best-performing XGBoost model achieved a patient-level accuracy of 86%. When the patient pool was further stratified, 96% accuracy was achieved within a tumor-stage-III sub-cohort.

Conclusions

ML-aided polarimetric analysis of collagenous stroma may provide prognostic value toward improving the clinical management of CRC patients.

Overview and Clinical Applications of Artificial Intelligence and Machine Learning in Cardiac Anesthesiology

Artificial intelligence- (AI) and machine learning (ML)-based applications are becoming increasingly pervasive in the healthcare setting. This has in turn challenged clinicians, hospital administrators, and health policymakers to understand such technologies and develop frameworks for safe and sustained clinical implementation. Within cardiac anesthesiology, challenges and opportunities for AI/ML to support patient care are presented by the vast amounts of electronic health data, which are collected rapidly, interpreted, and acted upon within the periprocedural area. To address such challenges and opportunities, in this article, the authors review 3 recent applications relevant to cardiac anesthesiology, including depth of anesthesia monitoring, operating room resource optimization, and transthoracic/transesophageal echocardiography, as conceptual examples to explore strengths and limitations of AI/ML within healthcare, and characterize this evolving landscape. Through reviewing such applications, the authors introduce basic AI/ML concepts and methodologies, as well as practical considerations and ethical concerns for initiating and maintaining safe clinical implementation of AI/ML-based algorithms for cardiac anesthesia patient care.

Whole Slide Imaging Technology and Its Applications: Current and Emerging Perspectives

Background. Whole slide imaging (WSI) represents a paradigm shift in pathology, serving as a necessary first step for a wide array of digital tools to enter the field. It utilizes virtual microscopy wherein glass slides are converted into digital slides and are viewed by pathologists by automated image analysis. Its impact on pathology workflow, reproducibility, dissemination of educational material, expansion of service to underprivileged areas, and institutional collaboration exemplifies a significant innovative movement. The recent US Food and Drug Administration approval to WSI for its use in primary surgical pathology diagnosis has opened opportunities for wider application of this technology in routine practice. Main Text. The ongoing technological advances in digital scanners, image visualization methods, and the integration of artificial intelligence-derived algorithms with these systems provide avenues to exploit its applications. Its benefits are innumerable such as ease of access through the internet, avoidance of physical storage space, and no risk of deterioration of staining quality or breakage of slides to name a few. Although the benefits of WSI to pathology practices are many, the complexities of implementation remain an obstacle to widespread adoption. Some barriers including the high cost, technical glitches, and most importantly professional hesitation to adopt a new technology have hindered its use in routine pathology. Conclusions. In this review, we summarize the technical aspects of WSI, its applications in diagnostic pathology, training, and research along with future perspectives. It also highlights improved understanding of the current challenges to implementation, as well as the benefits and successes of the technology. WSI provides a golden opportunity for pathologists to guide its evolution, standardization, and implementation to better acquaint them with the key aspects of this technology and its judicial use. Also, implementation of routine digital pathology is an extra step requiring resources which (currently) does not usually result increased efficiency or payment.

Use of artificial intelligence software in dental education: A study on assisted proximal caries assessment in bitewing radiographs

INTRODUCTION

Teaching of dental caries diagnostics is an essential part of dental education. Diagnosing proximal caries is a challenging task, and automated systems applying artificial intelligence (AI) have been introduced to assist in this respect. Thus, the implementation of AI for teaching purposes may be considered. The aim of this study was to assess the impact of an AI software on students' ability to detect enamel-only proximal caries in bitewing radiographs (BWs) and to assess whether proximal tooth overlap interferes with caries detection.

MATERIALS AND METHODS

The study included 74 dental students randomly allocated to either a test or control group. At two sessions, both groups assessed proximal enamel caries in BWs. At the first session, the test group registered caries in 25 BWs using AI software (AssistDent®) and the control group without using AI. One month later, both groups detected caries in another 25 BWs in a clinical setup without using the software. The student's registrations were compared with a reference standard. Positive agreement (caries) and negative agreement (no caries) were calculated, and t-tests were applied to assess whether the test and control groups performed differently. Moreover, t-tests were applied to test whether proximal overlap interfered with caries registration.

RESULTS

At the first and second sessions, 56 and 52 tooth surfaces, respectively, were detected with enamel-only caries according to the reference standard. At session 1, no significant difference between the control (48%) and the test (42%) group was found for positive agreement (p = .08), whereas the negative agreement was higher for the test group (86% vs. 80%; p = .02). At session 2, there was no significant difference between the groups. The test group improved for positive agreement from session 1 to session 2 (p < .001), while the control group improved for negative agreement (p < .001). Thirty-eight per cent of the tooth surfaces overlapped, and the mean positive agreement and negative agreement were significantly lower for overlapping surfaces than non-overlapping surfaces (p < .001) in both groups.

CONCLUSION

Training with the AI software did not impact on dental students' ability to detect proximal enamel caries in bitewing radiographs although the positive agreement improved over time. It was revealed that proximal tooth overlap interfered with caries detection.

Accuracy and reproducibility of ChatGPT's free version answers about endometriosis

OBJECTIVE

To evaluate the accuracy and reproducibility of ChatGPT's free version answers about endometriosis for the first time.

METHODS

Detailed internet searches to identify frequently asked questions (FAQs) about endometriosis have been performed. Scientific questions were prepared in accordance with the European Society of Human Reproduction and Embryology (ESHRE) endometriosis guidelines. An experienced gynecologist gave a score of 1-4 for each ChatGPT answer. The repeatability of ChatGPT answers about endometriosis was analyzed by asking each question twice, and the reproducibility of ChatGPT was accepted as scoring the answer to the same question in the same score category.

RESULTS

A total of 91.4% (n = 71) of all FAQs were answered completely, accurately, and sufficiently. ChatGPT had the highest accuracy in the symptom and diagnosis category (94.1%, 16/17 questions) and the lowest accuracy in the treatment category (81.3%, 13/16 questions). Furthermore, of the 40 questions based on the ESHRE endometriosis guidelines, 27 (67.5%) were classified as grade 1, seven (17.5%) as grade 2, and six (15.0%) as grade 3. The reproducibility rate of FAQs in the prevention, symptoms, and diagnosis, and complications categories was the highest (100% for all categories). The reproducibility rate was the lowest for questions based on the ESHRE endometriosis guidelines (70.0%).

CONCLUSION

ChatGPT accurately and satisfactorily responded to more than 90% of the questions about endometriosis, but to only 67.5% of questions based on the ESHRE endometriosis guidelines.

The Promise of Artificial Intelligence-Assisted Point-of-Care Ultrasonography in Perioperative Care

The role of point-of-care ultrasonography in the perioperative setting has expanded rapidly over recent years. Revolutionizing this technology further is integrating artificial intelligence to assist clinicians in optimizing images, identifying anomalies, performing automated measurements and calculations, and facilitating diagnoses. Artificial intelligence can increase point-of-care ultrasonography efficiency and accuracy, making it an even more valuable point-of-care tool. Given this topic's importance and ever-changing landscape, this review discusses the latest trends to serve as an introduction and update in this area.

Two artificial intelligence models underperform on examinations in a veterinary curriculum

OBJECTIVE

Advancements in artificial intelligence (AI) and large language models have rapidly generated new possibilities for education and knowledge dissemination in various domains. Currently, our understanding of the knowledge of these models, such as ChatGPT, in the medical and veterinary sciences is in its nascent stage. Educators are faced with an urgent need to better understand these models in order to unleash student potential, promote responsible use, and align AI models with educational goals and learning objectives. The objectives of this study were to evaluate the knowledge level and consistency of responses of 2 platforms of ChatGPT, namely GPT-3.5 and GPT-4.0.

SAMPLE

A total of 495 multiple-choice and true/false questions from 15 courses used in the assessment of third-year veterinary students at a single veterinary institution were included in this study.

METHODS

The questions were manually entered 3 times into each platform, and answers were recorded. These answers were then compared against those provided by the faculty members coordinating the courses.

RESULTS

GPT-3.5 achieved an overall performance score of 55%, whereas GPT-4.0 had a significantly (P < .05) greater performance score of 77%. Importantly, the performance scores of both platforms were significantly (P < .05) below that of the veterinary students (86%).

CLINICAL RELEVANCE

Findings of this study suggested that veterinary educators and veterinary students retrieving information from these AI-based platforms should do so with caution.

AI-guided histopathology predicts brain metastasis in lung cancer patients

Brain metastases can occur in nearly half of patients with early and locally advanced (stage I-III) non-small cell lung cancer (NSCLC). There are no reliable histopathologic or molecular means to identify those who are likely to develop brain metastases. We sought to determine if deep learning (DL) could be applied to routine H&E-stained primary tumor tissue sections from stage I-III NSCLC patients to predict the development of brain metastasis. Diagnostic slides from 158 patients with stage I-III NSCLC followed for at least 5 years for the development of brain metastases (Met+, 65 patients) versus no progression (Met-, 93 patients) were subjected to whole-slide imaging. Three separate iterations were performed by first selecting 118 cases (45 Met+, 73 Met-) to train and validate the DL algorithm, while 40 separate cases (20 Met+, 20 Met-) were used as the test set. The DL algorithm results were compared to a blinded review by four expert pathologists. The DL-based algorithm was able to distinguish the eventual development of brain metastases with an accuracy of 87% (p < 0.0001) compared with an average of 57.3% by the four pathologists and appears to be particularly useful in predicting brain metastases in stage I patients. The DL algorithm appears to focus on a complex set of histologic features. DL-based algorithms using routine H&E-stained slides may identify patients who are likely to develop brain metastases from those who will remain disease free over extended (>5 year) follow-up and may thus be spared systemic therapy. © 2024 The Authors. The Journal of Pathology published by John Wiley & Sons Ltd on behalf of The Pathological Society of Great Britain and Ireland.

Utility of artificial intelligence-based large language models in ophthalmic care

PURPOSE

With the introduction of ChatGPT, artificial intelligence (AI)-based large language models (LLMs) are rapidly becoming popular within the scientific community. They use natural language processing to generate human-like responses to queries. However, the application of LLMs and comparison of the abilities among different LLMs with their human counterparts in ophthalmic care remain under-reported.

RECENT FINDINGS

Hitherto, studies in eye care have demonstrated the utility of ChatGPT in generating patient information, clinical diagnosis and passing ophthalmology question-based examinations, among others. LLMs' performance (median accuracy, %) is influenced by factors such as the iteration, prompts utilised and the domain. Human expert (86%) demonstrated the highest proficiency in disease diagnosis, while ChatGPT-4 outperformed others in ophthalmology examinations (75.9%), symptom triaging (98%) and providing information and answering questions (84.6%). LLMs exhibited superior performance in general ophthalmology but reduced accuracy in ophthalmic subspecialties. Although AI-based LLMs like ChatGPT are deemed more efficient than their human counterparts, these AIs are constrained by their nonspecific and outdated training, no access to current knowledge, generation of plausible-sounding 'fake' responses or hallucinations, inability to process images, lack of critical literature analysis and ethical and copyright issues. A comprehensive evaluation of recently published studies is crucial to deepen understanding of LLMs and the potential of these AI-based LLMs.

SUMMARY

Ophthalmic care professionals should undertake a conservative approach when using AI, as human judgement remains essential for clinical decision-making and monitoring the accuracy of information. This review identified the ophthalmic applications and potential usages which need further exploration. With the advancement of LLMs, setting standards for benchmarking and promoting best practices is crucial. Potential clinical deployment requires the evaluation of these LLMs to move away from artificial settings, delve into clinical trials and determine their usefulness in the real world.

Evaluating the performance of large language models: ChatGPT and Google Bard in generating differential diagnoses in clinicopathological conferences of neurodegenerative disorders

This study explores the utility of the large language models (LLMs), specifically ChatGPT and Google Bard, in predicting neuropathologic diagnoses from clinical summaries. A total of 25 cases of neurodegenerative disorders presented at Mayo Clinic brain bank Clinico-Pathological Conferences were analyzed. The LLMs provided multiple pathologic diagnoses and their rationales, which were compared with the final clinical diagnoses made by physicians. ChatGPT-3.5, ChatGPT-4, and Google Bard correctly made primary diagnoses in 32%, 52%, and 40% of cases, respectively, while correct diagnoses were included in 76%, 84%, and 76% of cases, respectively. These findings highlight the potential of artificial intelligence tools like ChatGPT in neuropathology, suggesting they may facilitate more comprehensive discussions in clinicopathological conferences.

Artificial intelligence for cervical cancer screening: Scoping review, 2009-2022

BACKGROUND

The intersection of artificial intelligence (AI) with cancer research is increasing, and many of the advances have focused on the analysis of cancer images.

OBJECTIVES

To describe and synthesize the literature on the diagnostic accuracy of AI in early imaging diagnosis of cervical cancer following Preferred Reporting Items for Systematic Reviews and Meta-Analyses Extension for Scoping Reviews (PRISMA-ScR).

SEARCH STRATEGY

Arksey and O'Malley methodology was used and PubMed, Scopus, and Google Scholar databases were searched using a combination of English and Spanish keywords.

SELECTION CRITERIA

Identified titles and abstracts were screened to select original reports and cross-checked for overlap of cases.

DATA COLLECTION AND ANALYSIS

A descriptive summary was organized by the AI algorithm used, total of images analyzed, data source, clinical comparison criteria, and diagnosis performance.

MAIN RESULTS

We identified 32 studies published between 2009 and 2022. The primary sources of images were digital colposcopy, cervicography, and mobile devices. The machine learning/deep learning (DL) algorithms applied in the articles included support vector machine (SVM), random forest classifier, k-nearest neighbors, multilayer perceptron, C4.5, Naïve Bayes, AdaBoost, XGboots, conditional random fields, Bayes classifier, convolutional neural network (CNN; and variations), ResNet (several versions), YOLO+EfficientNetB0, and visual geometry group (VGG; several versions). SVM and DL methods (CNN, ResNet, VGG) showed the best diagnostic performances, with an accuracy of over 97%.

CONCLUSION

We concluded that the use of AI for cervical cancer screening has increased over the years, and some results (mainly from DL) are very promising. However, further research is necessary to validate these findings.

Biomaterial design for regenerating aged bone: materiobiological advances and paradigmatic shifts

China's aging demographic poses a challenge for treating prevalent bone diseases impacting life quality. As bone regeneration capacity diminishes with age due to cellular dysfunction and inflammation, advanced biomaterials-based approaches offer hope for aged bone regeneration. This review synthesizes materiobiology principles, focusing on biomaterials that target specific biological functions to restore tissue integrity. It covers strategies for stem cell manipulation, regulation of the inflammatory microenvironment, blood vessel regeneration, intervention in bone anabolism and catabolism, and nerve regulation. The review also explores molecular and cellular mechanisms underlying aged bone regeneration and proposes a database-driven design process for future biomaterial development. These insights may also guide therapies for other age-related conditions, contributing to the pursuit of 'healthy aging'.

Novel Technique for the Identification of Hip Implants Using Artificial Intelligence

BACKGROUND

The anticipated growth of total hip arthroplasty will result in an increased need for revision total hip arthroplasty. Preoperative planning, including identifying current implants, is critical for successful revision surgery. Artificial intelligence (AI) is promising for aiding clinical decision-making, including hip implant identification. However, previous studies have limitations such as small datasets, dissimilar stem designs, limited scalability, and the need for AI expertise. To address these limitations, we developed a novel technique to generate large datasets, tested radiographically similar stems, and demonstrated scalability utilizing a no-code machine learning solution.

METHODS

We trained, validated, and tested an automated machine learning-implemented convolutional neural network to classify 9 radiographically similar femoral implants with a metaphyseal-fitting wedge taper design. Our novel technique uses computed tomography-derived projections of a 3-dimensional scanned implant model superimposed within a computed tomography pelvis volume. We employed computer-aided design modeling and MATLAB to process and manipulate the images. This generated 27,020 images for training (22,957) and validation (4,063) sets. We obtained 786 test images from various sources. The performance of the model was evaluated by calculating sensitivity, specificity, and accuracy.

RESULTS

Our machine learning model discriminated the 9 implant models with a mean accuracy of 97.4%, sensitivity of 88.4%, and specificity of 98.5%.

CONCLUSIONS

Our novel hip implant detection technique accurately identified 9 radiographically similar implants. The method generates large datasets, is scalable, and can include historic or obscure implants. The no-code machine learning model demonstrates the feasibility of obtaining meaningful results without AI expertise, encouraging further research in this area.

Artificial intelligence for ultrasound scanning in regional anaesthesia: a scoping review of the evidence from multiple disciplines

BACKGROUND

Artificial intelligence (AI) for ultrasound scanning in regional anaesthesia is a rapidly developing interdisciplinary field. There is a risk that work could be undertaken in parallel by different elements of the community but with a lack of knowledge transfer between disciplines, leading to repetition and diverging methodologies. This scoping review aimed to identify and map the available literature on the accuracy and utility of AI systems for ultrasound scanning in regional anaesthesia.

METHODS

A literature search was conducted using Medline, Embase, CINAHL, IEEE Xplore, and ACM Digital Library. Clinical trial registries, a registry of doctoral theses, regulatory authority databases, and websites of learned societies in the field were searched. Online commercial sources were also reviewed.

RESULTS

In total, 13,014 sources were identified; 116 were included for full-text review. A marked change in AI techniques was noted in 2016-17, from which point on the predominant technique used was deep learning. Methods of evaluating accuracy are variable, meaning it is impossible to compare the performance of one model with another. Evaluations of utility are more comparable, but predominantly gained from the simulation setting with limited clinical data on efficacy or safety. Study methodology and reporting lack standardisation.

CONCLUSIONS

There is a lack of structure to the evaluation of accuracy and utility of AI for ultrasound scanning in regional anaesthesia, which hinders rigorous appraisal and clinical uptake. A framework for consistent evaluation is needed to inform model evaluation, allow comparison between approaches/models, and facilitate appropriate clinical adoption.

Towards precision medicine in breast imaging: A novel open mammography database with tailor-made 3D image retrieval for AI and teaching

This project addresses the global challenge of breast cancer, particularly in low-resource settings, by creating a pioneering mammography database. Breast cancer, identified by the World Health Organization as a leading cause of cancer death among women, often faces diagnostic and treatment resource constraints in low- and middle-income countries. To enhance early diagnosis and address educational setbacks, the project focuses on leveraging artificial intelligence (AI) technologies through a comprehensive database. Developed in collaboration with Ambra Health, a cloud-based medical image management software, the database comprises 941 mammography images from 100 anonymized cases, with 62 % including 3D images. Accessible through http://mamografia.unifesp.br, the platform facilitates a simple registration process and an advanced search system based on 169 clinical and imaging variables. The website, customizable to the user's native language, ensures data security through an automatic anonymization system. By providing high-resolution, 3D digital images and supplementary clinical information, the platform aims to promote education and research in breast cancer diagnosis, representing a significant advancement in resource-constrained healthcare environments.

Chat Generative Pretrained Transformer (ChatGPT) and Bard: Artificial Intelligence Does not yet Provide Clinically Supported Answers for Hip and Knee Osteoarthritis

BACKGROUND

Advancements in artificial intelligence (AI) have led to the creation of large language models (LLMs), such as Chat Generative Pretrained Transformer (ChatGPT) and Bard, that analyze online resources to synthesize responses to user queries. Despite their popularity, the accuracy of LLM responses to medical questions remains unknown. This study aimed to compare the responses of ChatGPT and Bard regarding treatments for hip and knee osteoarthritis with the American Academy of Orthopaedic Surgeons (AAOS) Evidence-Based Clinical Practice Guidelines (CPGs) recommendations.

METHODS

Both ChatGPT (Open AI) and Bard (Google) were queried regarding 20 treatments (10 for hip and 10 for knee osteoarthritis) from the AAOS CPGs. Responses were classified by 2 reviewers as being in "Concordance," "Discordance," or "No Concordance" with AAOS CPGs. A Cohen's Kappa coefficient was used to assess inter-rater reliability, and Chi-squared analyses were used to compare responses between LLMs.

RESULTS

Overall, ChatGPT and Bard provided responses that were concordant with the AAOS CPGs for 16 (80%) and 12 (60%) treatments, respectively. Notably, ChatGPT and Bard encouraged the use of non-recommended treatments in 30% and 60% of queries, respectively. There were no differences in performance when evaluating by joint or by recommended versus non-recommended treatments. Studies were referenced in 6 (30%) of the Bard responses and none (0%) of the ChatGPT responses. Of the 6 Bard responses, studies could only be identified for 1 (16.7%). Of the remaining, 2 (33.3%) responses cited studies in journals that did not exist, 2 (33.3%) cited studies that could not be found with the information given, and 1 (16.7%) provided links to unrelated studies.

CONCLUSIONS

Both ChatGPT and Bard do not consistently provide responses that align with the AAOS CPGs. Consequently, physicians and patients should temper expectations on the guidance AI platforms can currently provide.

Deep-Learning Automation of Preoperative Radiographic Parameters Associated With Early Periprosthetic Femur Fracture After Total Hip Arthroplasty

BACKGROUND

The radiographic assessment of bone morphology impacts implant selection and fixation type in total hip arthroplasty (THA) and is important to minimize the risk of periprosthetic femur fracture (PFF). We utilized a deep-learning algorithm to automate femoral radiographic parameters and determined which automated parameters were associated with early PFF.

METHODS

Radiographs from a publicly available database and from patients undergoing primary cementless THA at a high-volume institution (2016 to 2020) were obtained. A U-Net algorithm was trained to segment femoral landmarks for bone morphology parameter automation. Automated parameters were compared against that of a fellowship-trained surgeon and compared in an independent cohort of 100 patients who underwent THA (50 with early PFF and 50 controls matched by femoral component, age, sex, body mass index, and surgical approach).

RESULTS

On the independent cohort, the algorithm generated 1,710 unique measurements for 95 images (5% lesser trochanter identification failure) in 22 minutes. Medullary canal width, femoral cortex width, canal flare index, morphological cortical index, canal bone ratio, and canal calcar ratio had good-to-excellent correlation with surgeon measurements (Pearson's correlation coefficient: 0.76 to 0.96). Canal calcar ratios (0.43 ± 0.08 versus 0.40 ± 0.07) and canal bone ratios (0.39 ± 0.06 versus 0.36 ± 0.06) were higher (P < .05) in the PFF cohort when comparing the automated parameters.

CONCLUSIONS

Deep-learning automated parameters demonstrated differences in patients who had and did not have early PFF after cementless primary THA. This algorithm has the potential to complement and improve patient-specific PFF risk-prediction tools.

From technical to understandable: Artificial Intelligence Large Language Models improve the readability of knee radiology reports

PURPOSE

The purpose of this study was to evaluate the effectiveness of an Artificial Intelligence-Large Language Model (AI-LLM) at improving the readability of knee radiology reports.

METHODS

Reports of 100 knee X-rays, 100 knee computed tomography (CT) scans and 100 knee magnetic resonance imaging (MRI) scans were retrieved. The following prompt command was inserted into the AI-LLM: 'Explain this radiology report to a patient in layman's terms in the second person:[Report Text]'. The Flesch-Kincaid reading level (FKRL) score, Flesch reading ease (FRE) score and report length were calculated for the original radiology report and the AI-LLM generated report. Any 'hallucination' or inaccurate text produced by the AI-LLM-generated report was documented.

RESULTS

Statistically significant improvements in mean FKRL scores in the AI-LLM generated X-ray report (12.7 ± 1.0-7.2 ± 0.6), CT report (13.4 ± 1.0-7.5 ± 0.5) and MRI report (13.5 ± 0.9-7.5 ± 0.6) were observed. Statistically significant improvements in mean FRE scores in the AI-LLM generated X-ray report (39.5 ± 7.5-76.8 ± 5.1), CT report (27.3 ± 5.9-73.1 ± 5.6) and MRI report (26.8 ± 6.4-73.4 ± 5.0) were observed. Superior FKRL scores and FRE scores were observed in the AI-LLM-generated X-ray report compared to the AI-LLM-generated CT report and MRI report, p < 0.001. The hallucination rates in the AI-LLM generated X-ray report, CT report and MRI report were 2%, 5% and 5%, respectively.

CONCLUSIONS

This study highlights the promising use of AI-LLMs as an innovative, patient-centred strategy to improve the readability of knee radiology reports. The clinical relevance of this study is that an AI-LLM-generated knee radiology report may enhance patients' understanding of their imaging reports, potentially reducing the responder burden placed on the ordering physicians. However, due to the 'hallucinations' produced by the AI-LLM-generated report, the ordering physician must always engage in a collaborative discussion with the patient regarding both reports and the corresponding images.

LEVEL OF EVIDENCE

Level IV.

AI model disgorgement: Methods and choices

Over the past few years, machine learning models have significantly increased in size and complexity, especially in the area of generative AI such as large language models. These models require massive amounts of data and compute capacity to train, to the extent that concerns over the training data (such as protected or private content) cannot be practically addressed by retraining the model "from scratch" with the questionable data removed or altered. Furthermore, despite significant efforts and controls dedicated to ensuring that training corpora are properly curated and composed, the sheer volume required makes it infeasible to manually inspect each datum comprising a training corpus. One potential approach to training corpus data defects is model disgorgement, by which we broadly mean the elimination or reduction of not only any improperly used data, but also the effects of improperly used data on any component of an ML model. Model disgorgement techniques can be used to address a wide range of issues, such as reducing bias or toxicity, increasing fidelity, and ensuring responsible use of intellectual property. In this paper, we survey the landscape of model disgorgement methods and introduce a taxonomy of disgorgement techniques that are applicable to modern ML systems. In particular, we investigate the various meanings of "removing the effects" of data on the trained model in a way that does not require retraining from scratch.