Deep Convolutional Neural Network Self-Transfer Learning Medical Imaging
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional artificial intelligence techniques for digital image classification and analysis, such as medical imaging diagnosis, face challenges in accuracy and efficiency due to labor-intensive processes like pixel annotations and require extensive training data, making them inefficient for real-world applications.
Innovation Solution
A deep convolutional neural network with self-transfer learning employing sequential downsampling and upsampling in convolutional layers, using weakly supervised learning with image-level labels to classify and localize diseases in medical imaging data, improving the accuracy and efficiency of disease detection and visualization.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional AI techniques are used for digital image classification, then the system can perform basic classification, but the accuracy and efficiency are insufficient and labor-intensive processes are required
Solution Approach 1:
The system employs self-transfer learning where the deep convolutional neural network automatically learns from image-level labels without requiring pixel-level or voxel-level annotations. The network performs self-training by iteratively refining its own weights through downsampling and upsampling operations, eliminating the need for manual pixel annotations and achieving both high accuracy and efficiency
Solution Approach 2:
The patent applies iterative sequential downsampling and upsampling operations that dynamically change the spatial parameters of the input data. This parameter transformation allows the network to progressively refine its feature representations, improving classification accuracy while maintaining processing efficiency through automated parameter adjustment
2Reliability
If conventional AI techniques are used, then basic classification can be achieved, but extensive training data and labor-intensive annotations are required
Solution Approach 1:
The self-transfer learning mechanism enables the network to generate its own training data through iterative downsampling and upsampling operations. The network learns from a small set of image-level labels by automatically creating and refining its own feature representations, significantly reducing the quantity of training data needed while maintaining high reliability
Solution Approach 2:
The system performs preliminary downsampling operations that create coarser representations of the input data before final classification. This preliminary action allows the network to learn robust features from limited data, improving reliability without requiring extensive training data
3Measurement precision
If pixel annotations and voxel level annotations are used, then detailed localization can be achieved, but the process becomes labor-intensive and inefficient
Solution Approach 1:
The network automatically performs localization through its convolutional layers without requiring manual pixel or voxel annotations. The iterative downsampling and upsampling process enables the network to self-localize diseases and anatomical structures, achieving high precision while making the operation extremely easy (automated)
Data Source
AI summary
Systems and techniques for facilitating a deep convolutional neural network with self-transfer learning are presented. In one example, a system includes a machine learning component, a medical imaging diagnosis component and a visualization component. The machine learning component generates learned medical imaging output regarding an anatomical region based on a convolutional neural network that receives medical imaging data. The machine learning component also performs a plurality of sequential downsampling and upsampling of the medical imaging data associated with convolutional layers of the convolutional neural network. The medical imaging diagnosis component determines a classification and an associated localization for a portion of the anatomical region based on the learned medical imaging output associated with the convolutional neural network. The visualization component generates a multi-dimensional visualization associated with the classification and the localization for the portion of the anatomical region.