we present FM-IRD(also designated as Retina4IRD), the first foundation model-based AI system specifically designed for genotype diagnosis of IRD by emulating clinician decision-making reasoning process. Retina4IRD is capable of processing multimodal input including color fundus photography (CFP), optical coherence tomography (OCT), and descriptive medical metadata. Trained and validated on genetically confirmed cases from centers across China, South Korea, and Poland, the system generates a ranked list of candidate pathogenic genetic variants. Retina4IRD also can generated attention heatmaps to enhance decision-making interpretability. To validate its clinical impact, we conducted a prospective multicenter RCT involving 295 participants, rigorously assessing Retina4IRD's diagnostic accuracy and real-world utility. This will improve the standardized diagnosis of IRD diseases, effectively transforming the traditional time-consuming and resource-intensive diagnostic pathway into an efficient intelligent workflow
A randomized, controlled validation trial was conducted to evaluate the effectiveness of Retina4IRD (also designated as FM-IRD) in clinical settings. This trail was conducted at seven centers in China with large outpatient IRDs care . Participants were recruited from the outpatient of IRD clinics of the participating centers, or through referrals from collaborating retinal specialists. Given that genetic testing results were unavailable at enrollment, and to prioritize the model's ability to identify patients with actionable therapeutic targets, genotypes of our randomized controlled trial were categorized into 17 classes, including mutations with available gene therapies or ongoing clinical trials. All participants provided written informed consent before enrolment assessments. All participants subsequently underwent WES testing to confirm the causative genetic mutations, which served as the gold standard for IRD gene mutation diagnosis.
Study Type
INTERVENTIONAL
Allocation
RANDOMIZED
Purpose
OTHER
Masking
DOUBLE
Enrollment
300
A Foundation Model for Assisting the precision Diagnosis of Inherited Retinal Diseases: FM-IRDs
without FM-IRDs
Shanghai General Hospital, Shanghai Jiao Tong University
Shanghai, China
Shanghai general hospital
Shanghai, China
The diagnostic accuracy of IRDs
The top-5 gene mutation prediction accuracy, which will be assessed after all participants of the study have had WES test
Time frame: 21 days
Quality of the health management strategy
The secondary endpoint is defined as the quality of the health management strategy formulated for the top-one predicted genotype, which encompassed four dimensions: available interventions, recommended follow-up intervals, fertility counseling recommendations, and the necessity of multidisciplinary consultation. The quality of the strategy is evaluated using a comprehensive scoring system, which was developed by a panel of multiple experts. This system has a score range of 0 to 40, with higher scores indicating better outcomes. The final assessment is conducted independently by two senior IRD experts.
Time frame: 7 days
This platform is for informational purposes only and does not constitute medical advice. Always consult a qualified healthcare professional.