Shaping the future of myopia: artificial intelligence for vitreoretinal complications of high and pathologic myopia

Graefe s Archive for Clinical and Experimental Ophthalmology · 2026-02-04

PURPOSE: The global impact of myopia extends far beyond individual ocular health, posing significant challenges to healthcare systems worldwide. Artificial intelligence (AI), particularly deep learning (DL) applied to ophthalmic imaging, offers a promising strategy to ease constraints posed by the myopia epidemic by detecting subtle structural changes early. Here we describe the current literature on AI for detecting retinal sequelae of myopia, including retinal detachments (RD), myopic macular degeneration (MMD), and myopic traction maculopathy (MTM), with attention to imaging modality and model task (classification vs. segmentation). METHODS: A literature search was conducted to identify studies using DL to detect RD, MMD, and MTM across ophthalmic imaging modalities (including OCT and fundus photography, and where available fluorescein angiography and ultrasonography). RESULTS/FINDINGS: We reviewed 28 studies that piloted DL models usingclassification and/or segmentation approaches for RD (10 studies), MMD (12 studies), and MTM (6 studies). Reported performance for RD ranged from area under the curve (AUC) 86-100%, accuracy 79.3-98.9%, sensitivity 77.1-97.6%, and specificity 79.7-100%. For MMD, performance ranged from AUC 86-100%, accuracy 85.3-99.8%, sensitivity 37.1-97.8%, and specificity 91.5-99.9%. For MTM, performance ranged from AUC 93.8-99.7%, accuracy 94.3-99.3%, sensitivity 74.5-98.4%, and specificity 84.8-99.7%. Across studies, there was substantial heterogeneity in case definitions, datasets, and evaluation methods, and external validation was inconsistently reported. Many earlier studies used CNN-based architectures, while more recent work increasingly incorporates transformer-based backbones and pretrained or foundation models. CONCLUSION: Researchers have demonstrated excellent results for developing DL models that accurately classify and segment retinal pathologies associated with myopia. However, despite strong performance, additional work is needed to translate these models into clinical use, including robust external validation, calibration for clinical decision-making, and prospective evaluation, particularly for longitudinal prognostication of incident complications in pathologic myopia.

A generalizable deep learning system for cardiac MRI

Nature Biomedical Engineering · 2026-03-25

Cardiac MRI allows for a comprehensive assessment of myocardial structure, function and tissue characteristics. Here we describe a foundational vision system for cardiac MRI, capable of representing the breadth of human cardiovascular disease and health. Our deep-learning model is trained via self-supervised contrastive learning, in which visual concepts in cine-sequence cardiac MRI scans are learned from the raw text of the accompanying radiology reports. We train and evaluate our model on data from four large academic clinical institutions in the United States. We additionally showcase the performance of our models on the UK BioBank and two additional publicly available external datasets. We explore emergent capabilities of our system and demonstrate remarkable performance across a range of tasks, including the problem of left-ventricular ejection fraction regression and the diagnosis of 39 different conditions such as cardiac amyloidosis and hypertrophic cardiomyopathy. We show that our deep-learning system is capable of not only contextualizing the staggering complexity of human cardiovascular disease but can be directed towards clinical problems of interest, yielding impressive, clinical-grade diagnostic accuracy with a fraction of the training data typically required for such tasks.

A multimodal retinal aging clock for biological age prediction and systemic health assessment via OCT and fundus imaging

Scientific Reports · 2026-01-28

Herein we developed age clocks that predict biological age from fundus photography and optical coherence tomography. We evaluated our multimodal models' clinical relevance by examining their associations between predicted biological age and the Charlson Comorbidity Index (CCI). Study 1 assessed how models trained on normal eyes generalize to diseased eyes, and Study 2 tested whether incorporating disease labels improves performance and systemic associations. Models were fine-tuned to the imaging dataset to predict biological age. Linear regressors were trained on chronological and biological features to infer CCI. Gradient-weighted regression activation mapping also generated heatmaps to identify the model's region of focus. Prediction performance improved when trained on both normal and diseased eyes. Predicted biological age showed significantly stronger correlations with CCI than chronological age across both studies, supporting our algorithm's association with this validated measure of mortality. Thus, our algorithm may provide insight into systemic health burdens beyond that of traditional risk assessments.

A generalizable deep learning system for cardiac MRI

Nature Biomedical Engineering · 2026-03-25 · 3 citations

Cardiac MRI allows for a comprehensive assessment of myocardial structure, function and tissue characteristics. Here we describe a foundational vision system for cardiac MRI, capable of representing the breadth of human cardiovascular disease and health. Our deep-learning model is trained via self-supervised contrastive learning, in which visual concepts in cine-sequence cardiac MRI scans are learned from the raw text of the accompanying radiology reports. We train and evaluate our model on data from four large academic clinical institutions in the United States. We additionally showcase the performance of our models on the UK BioBank and two additional publicly available external datasets. We explore emergent capabilities of our system and demonstrate remarkable performance across a range of tasks, including the problem of left-ventricular ejection fraction regression and the diagnosis of 39 different conditions such as cardiac amyloidosis and hypertrophic cardiomyopathy. We show that our deep-learning system is capable of not only contextualizing the staggering complexity of human cardiovascular disease but can be directed towards clinical problems of interest, yielding impressive, clinical-grade diagnostic accuracy with a fraction of the training data typically required for such tasks.

The Effect of AI on the Radiologist Workforce: A Task-Based Analysis

medRxiv · 2025-12-22 · 1 citations

Background: The effect of AI algorithms on the radiology workforce has been a subject of commentary and controversy. There is now sufficient published evidence to support a quantitative task-based analysis to predict these effects. Purpose: To construct a quantitative, task-based model to predict the effect of AI on the radiology workforce using the best available evidence. Materials and Methods: We reviewed the literature to establish the tasks on which radiologists spend their time. We then developed categories of AI applications that could affect these tasks. We used published evidence to estimate the effect of each AI application on each radiology task using a 5-year time horizon. When published evidence was unavailable, we used our own judgment. Results: The model projects a 33% reduction in hours worked by radiologists in 5 years, with a range of 14% to 49%. The main effects are due to radiology report drafting for all modalities and study delegation for radiography and mammography. Conclusion: AI applications likely will cause a significant decrease in radiologist hours worked.. Given the relatively static radiology workforce and the continued growth in imaging volumes, radiologist job loss is unlikely for the foreseeable future.

STARC-9: A Large-scale Dataset for Multi-Class Tissue Classification for CRC Histopathology

PubMed · 2025-11-01

Multi-class tissue-type classification of colorectal cancer (CRC) histopathologic images is a significant step in the development of downstream machine learning models for diagnosis and treatment planning. However, publicly available CRC datasets used to build tissue classifiers often suffer from insufficient morphologic diversity, class imbalance, and low-quality image tiles, limiting downstream model performance and generalizability. To address this research gap, we introduce STARC-9 (STAnford coloRectal Cancer), a large-scale dataset for multi-class tissue classification. STARC-9 comprises 630,000 histopathologic image tiles uniformly sampled across nine clinically relevant tissue classes (each represented by 70,000 tiles), systematically extracted from hematoxylin & eosin-stained whole-slide images (WSI) from 200 CRC patients at the Stanford University School of Medicine. To construct STARC-9, we propose a novel framework, DeepCluster++, consisting of two primary steps to ensure diversity within each tissue class, followed by pathologist verification. First, an encoder from an autoencoder trained specifically on histopathologic images is used to extract feature vectors from all tiles within a given input WSI. Next, K-means clustering groups morphologically similar tiles, followed by an equal-frequency binning method to sample diverse patterns within each tissue class. Finally, the selected tiles are verified by expert gastrointestinal pathologists to ensure classification accuracy. This semi-automated approach significantly reduces the manual effort required for dataset curation while producing high-quality training examples. To validate the utility of STARC-9, we benchmarked baseline convolutional neural networks, transformers, and pathology-specific foundation models on downstream multi-class CRC tissue classification and segmentation tasks when trained on STARC-9 versus publicly available datasets, demonstrating superior generalizability of models trained on STARC-9. Although we demonstrate the utility of DeepCluster++ on CRC as a pilot use-case, it is a flexible framework that can be used for constructing high-quality datasets from large WSI repositories across a wide range of cancer and non-cancer applications. https://huggingface.co/datasets/Path2AI/STARC-9/tree/main https://github.com/Path2AI/STARC-9/.

Foundation versus domain-specific models for left ventricular segmentation on cardiac ultrasound

npj Digital Medicine · 2025-06-06 · 2 citations

The Segment Anything Model (SAM) was fine-tuned on the EchoNet-Dynamic dataset and evaluated on external transthoracic echocardiography (TTE) and Point-of-Care Ultrasound (POCUS) datasets from CAMUS (University Hospital of St Etienne) and Mayo Clinic (99 patients: 58 TTE, 41 POCUS). Fine-tuned SAM was superior or comparable to MedSAM. The fine-tuned SAM also outperformed EchoNet and U-Net models, demonstrating strong generalization, especially on apical 2-chamber (A2C) images (fine-tuned SAM vs. EchoNet: CAMUS-A2C: DSC 0.891 ± 0.040 vs. 0.752 ± 0.196, p < 0.0001) and POCUS (DSC 0.857 ± 0.047 vs. 0.667 ± 0.279, p < 0.0001). Additionally, SAM-enhanced workflow reduced annotation time by 50% (11.6 ± 4.5 sec vs. 5.7 ± 1.7 sec, p < 0.0001) while maintaining segmentation quality. We demonstrated an effective strategy for fine-tuning a vision foundation model for enhancing clinical workflow efficiency and supporting human-AI collaboration.

Expert-level validation of AI-generated medical text with scalable language models

Research Square · 2025-07-08 · 1 citations

PT1.01.01 Large Language Models to Extract Smoking History From Clinical Notes in EHR to Evaluate Lung Cancer Surveillance Strategies

Journal of Thoracic Oncology · 2025-10-01

From Detection to Mitigation: Addressing Bias in Deep Learning Models for Chest X-Ray Diagnosis

2025-12-01

Deep learning models have shown promise in improving diagnostic accuracy from chest X-rays, but they also risk perpetuating healthcare disparities when performance varies across demographic groups. In this work, we present a comprehensive bias detection and mitigation framework targeting sex, age, and race-based disparities when performing diagnostic tasks with chest X-rays. We extend a recent CNN-XGBoost pipeline to support multi-label classification and evaluate its performance across four medical conditions. We show that replacing the final layer of CNN with an eXtreme Gradient Boosting classifier improves the fairness of the subgroup while maintaining or improving the overall predictive performance. To validate its generalizability, we apply the method to different backbones, namely DenseNet-121 and ResNet-50, and achieve similarly strong performance and fairness outcomes, confirming its model-agnostic design. We further compare this lightweight adapter training method with traditional full-model training bias mitigation techniques, including adversarial training, reweighting, data augmentation, and active learning, and find that our approach offers competitive or superior bias reduction at a fraction of the computational cost. Finally, we show that combining eXtreme Gradient Boosting retraining with active learning yields the largest reduction in bias across all demographic subgroups, both in and out of distribution on the CheXpert and MIMIC datasets, establishing a practical and effective path toward equitable deep learning deployment in clinical radiology.