How Student Developers Used AI and Federated Learning to Revolutionize Eye Disease Diagnosis

Young developers from the University of Electronic Science and Technology of Chengdu created an AI‑driven eye‑disease auxiliary diagnosis system that won the China regional first‑place bronze award at the 2023 Huawei Developer Competition global finals.

Motivation and Team Formation

Sun Yang, a third‑year AI major, was inspired by a medical diagnosis project during the competition and assembled the “RongChuang YanYuan” team to apply neural networks, deep learning, and image recognition to address the lack of early eye‑health screening and uneven resource distribution.

Cloud Integration with Federated Learning

To protect patient privacy and avoid data silos, the team adopted a federated learning framework, keeping raw eye‑image data on local devices while uploading model parameters to the cloud. Using Huawei Cloud’s YunYao servers, they set up SSH connections, configured the necessary network and dependencies, and stored parameters in a cloud database for collaborative training, dramatically reducing setup time and cost.

With a Huawei‑provided Ascend Atlas 200I DK2 development board, they accelerated local model training before synchronizing results to the cloud.

Multimodal Data Fusion

Recognizing that clinicians need more than fundus images, the team fused fundus photos, fluorescein angiography (FFA), OCT images, and clinical measurements into weighted vectors fed into a neural network. Data augmentation and a custom hyper‑parameter‑tuned TransCNN model achieved 97.48% accuracy on 500 test cases, with a 3‑5% deviation from expert diagnoses.

The system also offers integrated image processing functions—vascular segmentation, classification, automatic annotation, and statistical reporting—to streamline doctors’ workflows.

Future Directions

The next goal is to convert 2D fundus images into 3D eye models, enable gesture‑based interaction, and incorporate Huawei Cloud’s voice‑recognition services, moving toward a “medical metaverse” platform for immersive diagnosis and surgical simulation.

Overall, the project demonstrates how student developers can leverage AI, cloud services, and federated learning to create high‑impact medical tools.