Intellectual Property India Publishes Patent Application for Multimodal Deep Learning Framework For Predicting Diabetic Retinopathy Blindness Progression

MUMBAI, India, July 24 -- Intellectual Property India has published a patent application (202641068355 A) filed by Aarupadai Veedu Institute Of Technology,vinayaka Missions Research Foundation on June 01, 2026, for Multimodal Deep Learning Framework For Predicting Diabetic Retinopathy Blindness Progression.

Inventors include Dhasaradhan K; and Jaichandran R.

The application for the patent was published on July 17, 2026, under issue no. 29/2026.

Abstract: Present invention generally relates to a system and method for predicting Diabetic Retinopathy(DR) blindness progression using cross attention fusion of retinal funds images and clinical data of DR patients. DR is the most common life threatening disease increasing worldwide. Accurate diagnosis and predicting of these diseases will help in providing better treatments. Traditional diagnosis methods take long time and requires skilled man power whose access in limited in rural areas. Alternatively, Machine Learning (ML) and Deep Learning (DL) methods can be used to assist ophthalmologist in diagnosis of DR. ML methods use DR patient’s clinical data and DL methods use DR patient’s fundus images in diagnoses of DR diseases. Most of the ML and DL methods use single d modality data an multimodal methods use simple concatenation of multimodality data in the diagnosis of DR diseases which may not be accurate. Hence these invention presents a Multimodal Deep Learning Frameworks using Cross Attention Fusion of Multimodality Data for Diagnosis and Prediction of Progression of DR. Present invention includes attention-based fundus image encoder that extracts feature from retinal fundus images, the MLP encoder that extracts features from DR patients' clinical data, and the cross-attention fusion module generates a feature map by associating features extracted from retinal fundus images and clinical data of DR patients and assigns more weightage to the most significant features. The Joint Diagnostic Network classifies the output as no DR, NPDR, and PDR and predicts the progression rate of DR as no progression, mild, moderate, or severe progression. Present invention is evaluated using metrics such as Accuracy, Precision, Sensitivity, Specificity, Fl-score and AUC-ROC. Results show the present invention has significant improvement in diagnosis and prediction of progression of DR compared to existing methods.

Disclaimer: Curated by HT Syndication.