We developed an artificial intelligence system, called QiLin, which was designed to assist anti-VEGF treatment decisions in retinal diseases. QiLin was trained and validated via over 20,000 optical coherence tomography images from multicenter datasets, demonstrating strong performance on both internal and external validation. To evaluate its real-world clinical utility, we conducted a randomized controlled trial that rigorously compares the accuracy of treatment decisions between a physician-only arm and an AI-assisted physician arm.
Study Type
INTERVENTIONAL
Allocation
RANDOMIZED
Purpose
OTHER
Masking
DOUBLE
Enrollment
200
A Comprehensive Deep Learning Model for Assisting the decision of anti-VEGF therapy: QiLin system
without QiLin assisted
Shanghai general hospital, Shanghai Jiao Tong University, Shanghai, 200080
Shanghai, China
Shanghai general hospital
Shanghai, China
Accuracy of the current anti-VEGF injection decision
The accuracy of the current anti-VEGF injection decision was defined as the proportion of injection decisions (yes or no) made by the physicians in the two arms that were in agreement with the independent senior expert.
Time frame: At enrollment
Accuracy of detecting active biomarkers on the current OCT image
The secondary endpoint was defined as the accuracy of detecting active biomarkers. For each patient, the physician was required to perform a binary classification (present vs. absent) for all of 8 pre-defined active biomarkers (PED, NV, IRF, SRF, SHRM, HRF, DRT or DME, and VMT), and was further confirmed by an independent senior retina specialist. The accuracy for per biomarker was calculated as the proportion of correct classifications for that biomarker, and then the average accuracy was calculated as the secondary endpoint.
Time frame: At enrollment
This platform is for informational purposes only and does not constitute medical advice. Always consult a qualified healthcare professional.