# Eye Disease Prediction using Ensemble Learning and Attention on OCT Scans

**Authors:** Gauri Naik, Nandini Narvekar, Dimple Agarwal, Nishita Nandanwar, Himangi Pande  
**arXiv:** [2311.15301](https://arxiv.org/abs/2311.15301)  
**Date:** 2023-11-26  
**Summary page:** https://www.summarizepaper.com/en/arxiv-id/2311.15301v1/

## Key points

- Comprehensive methodology utilizing machine learning, ensemble learning, and attention mechanisms for eye healthcare
- Advancements in technology combining machine learning and deep learning with OCT imaging
- Efficient detection of eye diseases from OCT images
- Classification of patients into disease-free or affected by specific conditions
- End-to-end web application for efficient eye disease prediction
- Segmentation using a trained custom UNet model
- Ensemble model consisting of InceptionV3 and Xception networks enhanced with self-attention layer
- Improved classification accuracy through self-attention approach
- Extensive experimentation and optimization for efficiency and optimal performance
- Effective eye disease prediction contributing to improved patient outcomes and proactive healthcare strategies
- Transformer model and StandAlone Self-Attention (SASA) module mentioned in related work

## Summary

, , , , 

This research paper presents a comprehensive methodology that utilizes machine learning, ensemble learning, and attention mechanisms to revolutionize eye healthcare. The advancements in technology, particularly the combination of machine learning and deep learning algorithms with Optical Coherent Technology (OCT) imaging, have opened new avenues for the detection and treatment of eye diseases. The proposed method focuses on efficient detection of eye diseases from OCT images and enables the classification of patients into disease-free (normal eyes) or affected by specific conditions such as Choroidal Neovascularization (CNV), Diabetic Macular Edema (DME), or Drusen. The research introduces an end-to-end web application that leverages machine learning and deep learning techniques for efficient eye disease prediction. Patients can submit their raw OCT scanned images through the application, which then undergo segmentation using a trained custom UNet model. The segmented images are fed into an ensemble model consisting of InceptionV3 and Xception networks enhanced with a self-attention layer. This self-attention approach utilizes the feature maps of individual models to achieve improved classification accuracy. The output of the ensemble model is aggregated to predict and classify various eye diseases. Extensive experimentation and optimization have been conducted to ensure the efficiency and optimal performance of the developed web application. The results demonstrate the effectiveness of the proposed approach in accurate eye disease prediction. By enabling early detection and accurate prediction of eye diseases, this methodology contributes to improved patient outcomes and proactive healthcare strategies. In related work, previous research has introduced the Transformer model, which utilizes self-attention mechanisms primarily for natural language processing tasks but has also been applied to image classification. Additionally, authors have presented the StandAlone Self-Attention (SASA) module designed specifically for image classification tasks, demonstrating its effectiveness when incorporated into vision models like ResNet and MobileNet. Overall, this research paper provides valuable insights into how machine learning, ensemble learning, and attention mechanisms can be harnessed to transform eye healthcare, offering potential for early detection, accurate prediction, and timely intervention in the treatment of eye diseases.