Retrospective cohort study finds that a fine-tuned model improves glaucoma detection, achieving 0.86 accuracy and 0.98 specificity in quality assessment.
Key Points
Model achieved 0.90 accuracy and 0.98 specificity for quality assessment, indicating high reliability in evaluating OCT scans.
Fine-tuned multimodal language model detected glaucoma with an accuracy of 0.86, highlighting its effectiveness for clinical use.
Analysis included 1,310 subjects' OCT scans, underscoring the extensive dataset for training and validation of the model.
High BLEU scores reflect strong alignment between generated text and reference clinical reports, supporting diagnostic accuracy.