Deep Learning Medical Image Segmentation via Data Augmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current automatic image segmentation technologies for abdominal and pelvic organs in medical images face challenges such as variability in shape and position, low contrast among adjacent organs, motion artifacts, and tissue changes, leading to inadequate segmentation algorithms.
Innovation Solution
A computer device and deep learning method using a processor and storage device to perform image pre-processing, augmentation, and training of an artificial intelligence model for medical image recognition, capable of segmenting images in 2D or 3D space based on CT or MRI scans, incorporating image pre-processing, augmentation, and post-processing to enhance segmentation accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional segmentation algorithms (threshold or edge segmentation) are used, then the processing speed is fast, but the segmentation accuracy is insufficient due to low contrast and variability in abdominal organs
Solution Approach 1:
The patent replaces conventional mechanical segmentation algorithms (threshold segmentation, edge detection) with a deep learning-based artificial intelligence model. This substitution enables the system to automatically learn complex patterns and features from medical images, achieving superior segmentation accuracy for abdominal organs despite the increased computational complexity of training and deploying neural networks.
Solution Approach 2:
The patent transforms the segmentation problem from using fixed manual parameters (thresholds, edge detection parameters) to using learned parameters from training data. The AI model automatically adapts its internal parameters (weights and biases) based on training with labeled medical images, enabling it to handle the variability in organ shape, size, and position that conventional fixed-parameter algorithms cannot address effectively.
2Reliability
If deep learning models are trained with limited data, then training time is reduced, but segmentation performance deteriorates due to insufficient learning of organ variations
Solution Approach 1:
The patent performs data augmentation and preprocessing operations before training the deep learning model. By preparing an enriched training dataset in advance through techniques such as geometric transformations, intensity modifications, and synthetic image generation, the system enables the model to learn more robust features that generalize better to unseen data, improving segmentation reliability without requiring excessively long training times.
3Manufacturing precision
If multiple medical image sequence slices are processed, then comprehensive 3D segmentation is achieved, but computational time increases significantly
Solution Approach 1:
The patent divides the 3D medical image into multiple 2D slices for independent processing by the deep learning model, then reconstructs the 3D segmentation results from these processed slices. This approach allows the system to leverage efficient 2D convolutional neural networks while still achieving comprehensive 3D segmentation, balancing computational efficiency with segmentation accuracy across the entire volume.
Data Source
AI summary
A deep learning method of an artificial intelligence model for medical image recognition is provided. The method includes the following steps: obtaining a first image set, where the first image set includes at least two images captured with different parameters; performing image pre-processing on each image of the first image set to obtain a second image set; performing image augmentation on the second image set to obtain a third image set; adding the third image set to a training image data set; and training the artificial intelligence model using the training image data set.