Medical Image Classification via Aggregated Cross Entropy Loss
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Differences in medical experience and labeling habits among doctors lead to inconsistencies, inaccuracies, and unlabeled medical images, making it difficult to classify and manage medical images effectively, particularly in CT scans.
Innovation Solution
An electronic device employs a 3D squeeze and excitation image classification model with cross entropy and aggregated cross entropy loss functions to automatically classify medical images by resampling, extracting features, and updating parameters using a backpropagation algorithm, enabling accurate phase identification of CT images.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If doctors manually label medical images based on their experience and habits, then the classification process is simple and quick, but the accuracy and consistency of labeling deteriorate due to subjective differences
Solution Approach 1:
The patent replaces the manual mechanical labeling process with an automated deep learning-based image classification model. The model automatically classifies medical images into different phases (non-contrast, arterial, venous, delay) without requiring manual doctor intervention, thereby eliminating subjective differences while maintaining high productivity.
Solution Approach 2:
The patent introduces an intermediary deep learning model that acts as a bridge between raw medical images and final classification results. This intermediary model processes images through multiple convolutional layers and attention mechanisms to produce consistent, accurate classifications, resolving the contradiction between speed and accuracy.
2Reliability
If multiple doctors label the same medical image, then more perspectives are obtained, but the time consumption and complexity of the classification process increase
Solution Approach 1:
The patent enables the medical image classification system to serve itself by automatically processing and classifying images without requiring multiple doctors' inputs. The deep learning model independently performs feature extraction, phase identification, and classification, achieving high reliability through its trained decision-making capability while eliminating time loss from multiple manual reviews.
3Device complexity
If traditional image classification models are used, then the model structure is simple and training is fast, but the classification accuracy and F1 score deteriorate on complex medical images
Solution Approach 1:
The patent segments the image classification task into multiple specialized components: initial convolutional layers for feature extraction, squeeze and excitation modules for attention mechanisms, and separate processing paths for different image phases. This segmentation allows each component to specialize in specific features, achieving high accuracy without overwhelming complexity.
Solution Approach 2:
The patent enhances the model by adding dimensional complexity through depthwise separable convolutions and attention mechanisms that operate in different feature spaces. These additional dimensions allow the model to capture complex patterns in medical images that simpler models miss, improving accuracy while managing computational complexity.
Data Source
AI summary
A medical image classification method such as CT (or CAT) scans includes receiving the CT scan or medical image, inputting the medical image into an image classification model, which provides a cross entropy (CE) loss function and an aggregated cross entropy (ACE) loss function. According to the ACE loss function, image samples with generic label are used as input data during model training. The medical image can be classified by using the image classification model, and a classification of the medical image is thereby obtained. The present disclosure can classify indeterminate or general medical images and even unlabeled images and thus realize supervision of medical data. A device for applying the method is also provided.