Multi-filter Auto-augmentation for Medical Image Classification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The limited availability of high-quality medical data for training deep learning-based computer-assisted diagnosis systems, particularly in gastroscopy, hinders the accuracy and efficiency of gastrointestinal cancer diagnosis due to the reliance on expert visual observation and the time-consuming process of data collection.
Innovation Solution
A method involving the training of multiple neural network models with different structures to auto-augment and filter medical image data, generating effective augmentation data, which is then used to train a second neural network model for improved classification accuracy, thereby enhancing the quality and quantity of training data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If multiple first neural network models with different structures are trained to classify medical images, then the quality of filtered augmentation data is improved, but the training time and computational resources increase
Solution Approach 1:
The patent applies preliminary action by pre-training multiple first neural network models with different structures before the main training phase. These pre-trained models are then used to filter and generate high-quality augmented training data through probability thresholding, which subsequently accelerates the training of the second neural network model by providing superior training inputs.
Solution Approach 2:
The patent introduces an intermediary mechanism where multiple first neural network models act as mediators between the raw medical image data and the second neural network model. These intermediary models generate probability distributions for augmented data, and data meeting predetermined probability thresholds are selected as effective training samples, thereby improving overall system performance.
2Reliability
If medical data collection follows strict approval procedures and patient consent requirements, then patient information protection is ensured, but the quantity of available training data decreases
Solution Approach 1:
The patent applies copying by generating multiple augmented copies of limited medical image data through data augmentation techniques. The system creates transformed versions of original images (rotations, flips, color adjustments) that serve as additional training samples, thereby expanding the effective training dataset size without requiring additional patient data collection.
Solution Approach 2:
The patent implements self-service by using the limited available medical data to train multiple first neural network models, which then automatically generate and filter augmented training data. This self-service mechanism allows the system to overcome data scarcity by leveraging its own trained models to create additional training examples, reducing dependence on extensive external data collection.
3Measurement precision
If expert visual observation is used for gastrointestinal cancer diagnosis, then diagnostic accuracy can be achieved, but the process is time-consuming and subject to expert fatigue and skill variations
Solution Approach 1:
The patent replaces the mechanical system of expert visual observation with an automated neural network-based classification system. The trained second neural network model processes medical images through automated feature extraction and classification, eliminating human factors such as fatigue and skill variation while maintaining high diagnostic accuracy and significantly reducing diagnosis time.
Solution Approach 2:
The patent incorporates feedback mechanisms where the neural network models process medical images and provide classification results with associated probability distributions. The system uses predetermined probability thresholds to validate predictions, and the feedback from multiple first models informs the training and decision-making of the second model, creating a robust automated diagnostic loop.
Data Source
AI summary
A method for training a medical image classification model using a multi-filter auto-augmentation includes training, by using a training dataset including raw medical image data, a plurality of first neural network models to classify medical image data into a predetermined class, in which the plurality of first neural network models have different neural network model structures, auto-augmenting the raw medical image data to generate medical image augmentation data, filtering data of the medical image augmentation data, which has a class probability of belonging to a class classified by each of the plurality of first neural network models, equal to or greater than a predetermined criterion, as effective augmentation data, and training, by using a training dataset including the effective augmentation data and the raw medical image data, a second neural network model to classify medical image data into a predetermined class.