Background: We present BODHI (Balanced, Open-minded, Diagnostic, Humble, and Inquisitive), an engineering framework for curiosity-driven and humble clinical decision support AI. Despite growing capabilities, large language models (LLMs) often express inappropriate confidence, conflating statistical pattern recognition with genuine medical understanding. BODHI addresses this through a dual-reflective architecture that: (1) decomposes epistemic uncertainty into task-specific dimensions, and (2) constrains model responses using virtue-based stance rules derived from a Virtue Activation Matrix. Methods: We validate the framework through controlled evaluation on 200 clinical vignettes from HealthBench Hard, assessing GPT-4o-mini and GPT-4.1-mini across 5 random seeds (1,800 total observations). Statistical analysis included bootstrap resampling, paired t-tests, and effect size computation (Supplementary Materials S3). Findings: BODHI significantly improved overall clinical response quality (GPT-4.1-mini: +17.3pp, p<0.0001, Cohen's d= 0.50; GPT-4o-mini: +7.4pp, p<0.0001, Cohen's d= 0.22) while achieving very large effect sizes on curiosity (context-seeking rate: Cohen's d = 16.38 and 19.54) and humility (hedging: d = 5.80 for GPT-4.1-mini) metrics. Crucially, 97.3% of GPT-4.1-mini responses and 73.5% of GPT-4o-mini responses included appropriate clarifying questions, compared to 7.8% and 0.0% at baseline, demonstrating the framework's effectiveness in eliciting information-gathering behavior. Interpretation: These findings suggest LLMs can be reliably constrained to operate within epistemic boundaries when provided with structured uncertainty decomposition and virtue-aligned response rules, offering a pathway toward safer clinical AI deployment. Keywords: Human-AI Collaboration, Epistemic Humility, Curiosity-Driven Learning, Healthcare AI, Uncertainty Quantification, Clinical Decision Support ### Competing Interest Statement The authors have declared no competing interest. ### Funding Statement This study did not receive any funding. ### Author Declarations I confirm all relevant ethical guidelines have been followed, and any necessary IRB and/or ethics committee approvals have been obtained. Yes I confirm that all necessary patient/participant consent has been obtained and the appropriate institutional forms have been archived, and that any patient/participant/sample identifiers included were not known to anyone (e.g., hospital staff, patients or participants themselves) outside the research group so cannot be used to identify individuals. Yes I understand that all clinical trials and any other prospective interventional studies must be registered with an ICMJE-approved registry, such as ClinicalTrials.gov. I confirm that any such study reported in the manuscript has been registered and the trial registration ID is provided (note: if posting a prospective study registered retrospectively, please provide a statement in the trial ID field explaining why the study was not registered in advance). Yes I have followed all appropriate research reporting guidelines, such as any relevant EQUATOR Network research reporting checklist(s) and other pertinent material, if applicable. Yes All data are available at github.com/sebasmos/humbleai-healthbench and https://pypi.org/project/bodhi-llm/

The black-box nature of deep learning (DL) models presents a significant challenge for their adoption in clinical settings. The field of explainable artificial intelligence (XAI) has emerged to improve the transparency and interpretability of models. However, current techniques do not adequately describe the reasoning underpinning DL models. This study replicates and extends previous research on the use of texture analysis to improve interpretability in clinically geared segmentation tasks. We evaluate Law’s Texture Energy Measures (LTEMs) in the learning and decision-making processes of different DL architectures. We extend the work to include breast cancer, skin lesion, and gastrointestinal polyp datasets, as well as CLAHE-enhanced datasets to identify any divergence in learning. Experimental results reiterate that LTEMs, specifically level-edge convolution masks, are highly influential across multiple DL architectures. Additionally, Gray-Level Co-occurrence Matrix (GLCM) analysis highlights autocorrelation as a key descriptor. The results confirm that texture-based representations, learned primarily in the early layers of the network, are sufficient for robust learning. Through LTEMs, we can characterize the patterns learned in DL and associate these patterns with verbal descriptions and clinically objective measures, thus translating the DL learning into human terms. This psychophysical approach eases the clinical interpretability of DL models. Code availability: https://github.com/xrai-lib/xai-texture.

Multimodal learning has witnessed remarkable advancements in recent years, particularly with the integration of attention-based models, leading to significant performance gains across a variety of tasks. Parallel to this progress, the demand for explainable artificial intelligence (XAI) has spurred a growing body of research aimed at interpreting the complex decision-making processes of these models. This systematic literature review analyzes research published between January 2020 and early 2024 that focuses on the explainability of multimodal models. Framed within the broader goals of XAI, we examine the literature across multiple dimensions, including model architecture, modalities involved, explanation algorithms and evaluation methodologies. Our analysis reveals that most studies are concentrated on vision-language and language-only models, with attention-based techniques being the most commonly employed for explanation. However, these methods often fall short in capturing the full spectrum of interactions between modalities, a challenge further compounded by the architectural heterogeneity across domains. Importantly, we find that evaluation methods for XAI in multimodal settings are largely non-systematic, lacking consistency, robustness, and consideration for modality-specific cognitive and contextual factors. To address these gaps, we not only synthesize findings from the surveyed works but also incorporate a complementary analysis that integrates recent and emerging advances driving multimodal explainability. Based on these insights, we provide a comprehensive set of recommendations aimed at promoting rigorous, transparent, and standardized evaluation and reporting practices in multimodal XAI research. Our goal is to support future research in more interpretable, accountable, and responsible multimodal AI systems, with explainability at their core.

Imbalanced datasets pose a considerable challenge in training deep learning (DL) models for medical diagnostics, particularly for segmentation tasks. Imbalance may be associated with annotation quality limited annotated datasets, rare cases, or small-scale regions of interest (ROIs). These conditions adversely affect model training and performance, leading to segmentation boundaries which deviate from the true ROIs. Traditional loss functions, such as Binary Cross Entropy, replicate annotation biases and limit model generalization. We propose a novel, statistically driven, conditionally adaptive loss function (CALF) tailored to accommodate the conditions of imbalanced datasets in DL training. It employs a data-driven methodology by estimating imbalance severity using statistical methods of skewness and kurtosis, then applies an appropriate transformation to balance the training dataset while preserving data heterogeneity. This transformative approach integrates a multifaceted process, encompassing preprocessing, dataset filtering, and dynamic loss selection to achieve optimal outcomes. We benchmark our method against conventional loss functions using qualitative and quantitative evaluations. Experiments using large-scale open-source datasets (i.e., UPENN-GBM, UCSF, LGG, and BraTS) validate our approach, demonstrating substantial segmentation improvements. Code availability: https://anonymous.4open.science/r/MICCAI-Submission-43F9/.

Motivation: Medical researchers obtain knowledge about the prevention and treatment of disability and disease using physical measurements and image data. To assist in this endeavor, feature extraction packages are available that are designed to collect data from the image structure. While these packages have found prominence with their application in the medical domain, several key features, particularly as related to shape, are not as readily accessible. In this study, we aim to augment current works by creating a dedicated package for shape-based features. The significance of shape-based features has been explored extensively in research for several decades, but there is no single package available in which all shape-related features can be extracted easily by the researcher. Results: PyCellMech has been crafted to address this gap. The PyCellMech package extracts three classes of shape features, which are classified as one-dimensional, geometric, and polygonal. Features can easily be extracted using the command line interface (CLI) and include both single- and multi-class feature extraction. Features are extracted quantitatively in CSV form as well as qualitatively where the features can be visualized against the original image. The open-source package is readily available for installation as a Python package. Future iterations will be expanded to include other feature classes, such as scale-space, and GUI-based versions to accommodate users who prefer not to work with CLIs. Availability: PyCellMech is freely available at https://github.com/icm-dac/pycellmech or alternatively can be installed in Python using pip install pycellmech.

Background: Several studies have investigated various features and models in order to understand the growth and progression of the ocular disease geographic atrophy (GA). Commonly assessed features include age, sex, smoking, alcohol consumption, sedentary lifestyle, hypertension, and diabetes. There have been inconsistencies regarding which features correlate with GA progression. Chief amongst these inconsistencies is whether the investigated features are readily available for analysis across various ophthalmic institutions. Methods:In this study, we focused our attention on the association of fundus autofluorescence (FAF) imaging features and GA progression. Our method included feature extraction using radiomic processes and feature ranking by machine learning incorporating the algorithm XGBoost to determine the best-ranked features. This led to the development of an image-based linear mixed-effects model, which was designed to account for slope change based on within-subject variability and inter-eye correlation. Metrics used to assess the linear mixed-effects model included marginal and conditional R2, Pearson’s correlation coefficient (r), root mean square error (RMSE), mean error (ME), mean absolute error (MAE), mean absolute deviation (MAD), the Akaike Information Criterion (AIC), the Bayesian Information Criterion (BIC), and loglikelihood. Results: We developed a linear mixed-effects model with 15 image-based features. The model results were as follows: R2 = 0.96, r = 0.981, RMSE = 1.32, ME = −7.3 × 10−15, MAE = 0.94, MAD = 0.999, AIC = 2084.93, BIC = 2169.97, and log likelihood = −1022.46. Conclusions: The advantage of our method is that it relies on the inherent properties of the image itself, rather than the availability of clinical or demographic data. Thus, the image features discovered in this study are universally and readily available across the board.

The human brain receives nutrients and oxygen through an intricate network of blood vessels. Pathology affecting small vessels, at the mesoscopic scale, represents a critical vulnerability within the cerebral blood supply and can lead to severe conditions, such as Cerebral Small Vessel Diseases. The advent of 7 Tesla MRI systems has enabled the acquisition of higher spatial resolution images, making it possible to visualise such vessels in the brain. However, the lack of publicly available annotated datasets has impeded the development of robust, machine learning-driven segmentation algorithms. To address this, the SMILE-UHURA challenge was organised. This challenge, held in conjunction with the ISBI 2023, in Cartagena de Indias, Colombia, aimed to provide a platform for researchers working on related topics. The SMILE-UHURA challenge addresses the gap in publicly available annotated datasets by providing an annotated dataset of Time-of-Flight angiography acquired with 7T MRI. This dataset was created through a combination of automated pre-segmentation and extensive manual refinement. In this manuscript, sixteen submitted methods and two baseline methods are compared both quantitatively and qualitatively on two different datasets: held-out test MRAs from the same dataset as the training data (with labels kept secret) and a separate 7T ToF MRA dataset where both input volumes and labels are kept secret. The results demonstrate that most of the submitted deep learning methods, trained on the provided training dataset, achieved reliable segmentation performance. Dice scores reached up to 0.838 ± 0.066 and 0.716 ± 0.125 on the respective datasets, with an average performance of up to 0.804 ± 0.15.

OBJECTIVE:To examine the nature and severity of badminton-related ocular injuries in Melbourne, Australia. METHODS:This is a retrospective chart review. A search of the medical records was conducted for patients presenting to the ED at The Royal Victorian Eye and Ear Hospital, with badminton-related eye injuries from June 2018 to May 2023. Data were extracted, focusing on injury mechanism, patient demographics and treatment outcomes. RESULTS:In total, 88 patients were included in the study, comprising 64 (73%) men and 24 (27%) women. The mean patient age was 36.13 years. The most common injury was hyphaema (73%), followed by commotio retinae (45%). One patient sustained a penetrating eye injury when a shuttlecock shattered the spectacles he was wearing during play. Medical intervention was required for 90% of patients. The most common interventions were steroid eye drops (80%) and cycloplegic eyedrops (76%). A total of six (7%) patients required surgical management. For those 69 patients followed up at The Royal Victorian Eye and Ear Hospital, 77% of patients had a final best-corrected visual acuity of 6/6 or greater. CONCLUSIONS:Hyphaema, commotio retinae and traumatic uveitis were the most commonly diagnosed injuries. The majority of patients with badminton-related eye injuries required medical treatment, and some necessitated surgical intervention. To mitigate these risks, there is a pressing need to develop an eye safety policy for Australian badminton players, and players should exercise caution when wearing spectacles during play to prevent potential penetrating eye injuries.

Objective. The distribution of hypoxia within tissues plays a critical role in tumor diagnosis and prognosis. Recognizing the significance of tumor oxygenation and hypoxia gradients, we introduce mathematical frameworks grounded in mechanistic modeling approaches for their quantitative assessment within a tumor microenvironment. By utilizing known blood vasculature, we aim to predict hypoxia levels across different tumor types. Approach. Our approach offers a computational method to measure and predict hypoxia using known blood vasculature. By formulating a reaction-diffusion model for oxygen distribution, we derive the corresponding hypoxia profile. Main results. The framework successfully replicates observed inter- and intra-tumor heterogeneity in experimentally obtained hypoxia profiles across various tumor types (breast, ovarian, pancreatic). Additionally, we propose a data-driven method to deduce partial differential equation models with spatially dependent parameters, which allows us to comprehend the variability of hypoxia profiles within tissues. The versatility of our framework lies in capturing diverse and dynamic behaviors of tumor oxygenation, as well as categorizing states of vascularization based on the dynamics of oxygen molecules, as identified by the model parameters. Significance. The proposed data-informed mechanistic method quantitatively assesses hypoxia in the tumor microenvironment by integrating diverse histopathological data and making predictions across different types of data. The framework provides valuable insights from both modeling and biological perspectives, advancing our comprehension of spatio-temporal dynamics of tumor oxygenation.

Summary: Medical researchers obtain knowledge about the prevention and treatment of disability and disease using physical measurements and image data. To assist in this endeavor, feature extraction packages are available that are designed to collect data from the image structure. In this study, we aim to augment current works by adding to the current mix of shape-based features. The significance of shape-based features has been explored extensively in research for several decades, but there is no single package available in which all shape-related features can be extracted easily by the researcher. PyCellMech has been crafted to address this gap. The PyCellMech package extracts three classes of shape features, which are classified as one-dimensional, geometric, and polygonal. Future iterations will be expanded to include other feature classes, such as scale-space. Availability and implementation: PyCellMech is freely available at https://github.com/icm-dac/pycellmech.

There has been a long history of women innovators producing outstanding contributions to society and public benefit yet having their work passed over or sidelined or attributed to male colleagues. This phenomenon has been coined the Matilda Effect. The amendments to the record of human achievements are now taking place, with an increasing pace in recent times due to greater social enlightenment and awareness and the interest in social justice. However, there remains a gap that must be addressed. In this article, we demonstrate the disparity in scientific recognition for a handful of case studies through search data collected via Google Trends and plotted as time-series figures and choropleth maps. Search trends reflect a noticeable divergence between recognition of female and male innovators. However, we note that in more well-known cases of the Matilda Effect, such as the historical account of Rosalind Franklin vs. James Watson and Francis Crick, the differences become less pronounced, emphasizing the importance of publicizing recognition. In response to this revelation, this article presents the stories of several women innovators and their great achievements. We identify the truth behind several discoveries and inventions, while revealing the full nature of this historical problem of social exclusion.

The build-up of lipofuscin—an age-associated biomarker referred to as hyperfluorescence—is considered a precursor in the progression of geographic atrophy (GA). Prior studies have attempted to classify hyperfluorescent regions to explain varying rates of GA progression. In this study, digital image processing and unsupervised learning were used to (1) completely automate the extraction of hyperfluorescent regions from images, and (2) evaluate prospective patterns and groupings of hyperfluorescent areas associated with varying levels of GA progression. Patterns were determined by clustering methods, such as k-Means, and performance was evaluated using metrics such as the Silhouette Coefficient (SC), the Davies–Bouldin Index (DBI), and the Calinski–Harabasz Index (CHI). Automated extraction of hyperfluorescent regions was carried out using pseudocoloring techniques. The approach revealed three distinct types of hyperfluorescence based on color intensity changes: early-stage hyperfluorescence, intermediate-stage hyperfluorescence, and late-stage hyperfluorescence, with the early and late stages having three additional subclassifications that could explain varying levels of GA progression. The performance metrics for early-stage hyperfluorescence were SC = 0.597, DBI = 0.915, and CHI = 186.989. For late-stage hyperfluorescence, SC = 0.593, DBI = 1.013, and CHI = 217.325. No meaningful subclusters were identified for the intermediate-stage hyperfluorescence, possibly because it is a transitional phase of hyperfluorescence progression.

Our website uses cookies to enhance your experience. By continuing to use our site, or clicking "Continue," you are agreeing to our Cookie Policy | Continue JAMA Ophthalmology HomeNew OnlineCurrent IssueFor Authors Podcast Journals JAMA JAMA Network Open JAMA Cardiology JAMA Dermatology JAMA Health Forum JAMA Internal Medicine JAMA Neurology JAMA Oncology JAMA Ophthalmology JAMA Otolaryngology–Head & Neck Surgery JAMA Pediatrics JAMA Psychiatry JAMA Surgery Archives of Neurology & Psychiatry (1919-1959) JN Learning / CMESubscribeJobsInstitutions / LibrariansReprints & Permissions Terms of Use | Privacy Policy | Accessibility Statement 2023 American Medical Association. All Rights Reserved Search All JAMA JAMA Network Open JAMA Cardiology JAMA Dermatology JAMA Forum Archive JAMA Health Forum JAMA Internal Medicine JAMA Neurology JAMA Oncology JAMA Ophthalmology JAMA Otolaryngology–Head & Neck Surgery JAMA Pediatrics JAMA Psychiatry JAMA Surgery Archives of Neurology & Psychiatry Input Search Term Sign In Individual Sign In Sign inCreate an Account Access through your institution Sign In Purchase Options: Buy this article Rent this article Subscribe to the JAMA Ophthalmology journal

Cutaneous melanoma is an invasive cancer with a worldwide annual death toll of 57,000 (Arnold et al., JAMA Dermatol 2022). In a metastatic state, surgical interventions are not curative and must be coupled with targeted therapy, or immunotherapy. However, resistance appears almost systematically and late-stage prognosis can remain poor. The complexity to eradicate melanoma stems from its plasticity; these cancer cells continually adapt to the tumor microenvironment, which leads to treatment resistance. Our primary assumption is that therapeutic resistance relies in part on a series of non-genetic transitions including changes in the metabolic states of these cancer cells. The 3D spatial distribution of blood vessels that are sources of nutrition and oxygen that drive this metabolic status is an important variable for understanding zoning aspects of this adaptation process. Using Whole Slide Images (WSI) of melanoma tumors from Patient-Derived Xenograft (PDX) mouse models, we build 3D vascular models to help predict and understand the metabolic states of cancer cells within the tumor. Our 3D reconstruction pipeline was based on PDX tumor samples sectioned over 2mm depth and stained with Hematoxylin and Eosin (H&E). The pipeline involves three primary steps, including 2D vessel segmentation using Deep Learning, intensity- and affine-based image registration, and 3D reconstruction using interpolation and 3D rendering (allowing for better interaction with biologists, pathologists, and clinicians). The originality of our computer-assisted pipeline is its capability to (a) deal with sparse data (i.e., not all tissue sections were readily available), and (b) adapt to a multitude of WSI-related challenges (e.g., epistemic uncertainty, extended processing times due to WSI scale, etc.). We posit both our 3D reconstruction pipeline, quantitative results of the major stages of the process, and a detailed illustration of the challenges faced, presenting resolutions to improve the pipeline's efficiency.

Exploratory data analysis and statistical moments were used to investigate the potential impact of ceiling and floor effects in medical trials. A total of 150 treatment-naive eyes were assessed in a retrospective case study of patients who were treated with anti-VEGF injections for wet age-related macular degeneration. The experimental results revealed that ceiling and floor effects are problematic in data analysis and may result in serious errors when using standard parametric tests. The case study provided insights relating to methodology in medical trials, experimental data analysis, and statistical inference, as applied to the interpretation of treatment response limits. Suggestions are provided for statistical data pre-processing and post-processing when significantly skewed distributions are present in response groups.