Multi-Disease Image Classification via Residual Network Merging
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional disease detection and classification models can only detect a single disease, requiring multiple models and training images for each disease, which is time-consuming and inefficient.
Innovation Solution
A computer-implemented method using a residual network with N number of residual blocks, (N−1) pooling layers, and (N−1) convolutional layers to classify images, processing outputs through convolutional layers, vectorizing, and concatenating them for classification, enabling detection of multiple diseases with a single model.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If multiple separate classification models are used to detect different diseases, then detection coverage for multiple diseases is improved, but device complexity and training time increase significantly
Solution Approach 1:
The patent merges multiple disease detection functions into a single unified classification model. The model architecture integrates multiple classification heads, each responsible for detecting different diseases (e.g., diabetic retinopathy, glaucoma, hypertensive retinopathy), allowing one model to perform what previously required multiple separate models. This reduces system complexity while maintaining comprehensive disease detection coverage.
Solution Approach 2:
The classification model is designed with universal functionality to detect multiple types of diseases simultaneously. By incorporating multiple classification heads and using a shared feature extraction backbone, the model achieves multi-functionality, eliminating the need for separate specialized models for each disease type while improving detection coverage across multiple conditions.
2Reliability
If multiple separate classification models are used for each disease, then detection accuracy for each disease can be optimized, but training time and computational resources increase
Solution Approach 1:
The patent combines multiple disease detection tasks into a single training process. By sharing the feature extraction backbone across multiple classification heads, the model learns common features once and applies them to detect different diseases, significantly reducing training time compared to training separate models for each disease while maintaining or improving detection accuracy through joint optimization.
Solution Approach 2:
The model performs preliminary feature extraction that is reusable across multiple disease detection tasks. By pre-processing and extracting common features once through the shared backbone, the system avoids redundant computation during both training and inference, reducing overall computational time and resources while preserving disease-specific detection accuracy through specialized classification heads.
3Measurement precision
If multiple separate models are trained for different diseases, then each model can be specialized, but the quantity of training data required increases
Solution Approach 1:
The patent merges multiple disease classification tasks into a single multi-task learning framework. By training one model on combined datasets from multiple disease conditions, the system achieves specialized detection precision for each disease while using a unified training process that reduces the total volume of training data required compared to training separate specialized models for each disease.
Data Source
AI summary
A computer-implemented method is provided. The computer-implemented method includes classifying an image using a classification model having a residual network. Classifying the image using the classification model includes inputting an input image into the residual network having N number of residual blocks sequentially connected, N≥2, (N−1) number of pooling layers respectively between two adjacent residual blocks of the N number of residual blocks, and (N−1) number of convolutional layers respectively connected to first to (N−1)-th residual blocks of the N number of residual blocks; processing outputs from the first to the (N−1)-th residual blocks of the N number of residual blocks respectively through the (N−1) number of convolutional layers; vectorizing outputs respectively from the (N−1) number of convolutional layers to generate (N−1) number of vectorized outputs; vectorizing an output from a last residual block of the N number of residual blocks to generate a last vectorized output.