Machine Learning Device for Tomographic Image Data Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing machine learning techniques for processing three-dimensional medical image data, such as CT images, face challenges when dividing data for memory constraints, leading to adverse effects on learning accuracy when only a part of an organ is included in the divided data.
Innovation Solution
A machine learning device and method that divides learning data into subsets, discriminates regions to be excluded from the learning process based on volume thresholds, and performs machine learning on regions other than the excluded areas, using techniques like Intersection over Union (IoU) for accuracy calculation and re-dividing data to ensure complete organ inclusion.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If volume data is divided for memory constraints, then memory usage is reduced, but learning accuracy deteriorates when only a part of an organ is included in the divided data
Solution Approach 1:
The patent segments the volume data into multiple smaller datasets along the axial direction, allowing the system to process large medical images while maintaining memory efficiency. Each segmented dataset is processed independently through the machine learning model, and results are aggregated to produce final organ detection outputs.
Solution Approach 2:
The patent performs preliminary actions by pre-processing the volume data to identify and extract complete organs before segmentation. This ensures that when data is divided, complete organ structures are preserved in at least one segment, preventing the loss of critical anatomical information that would degrade learning accuracy.
2Productivity
If data is divided into smaller subsets, then processing efficiency is improved, but detection accuracy of organs with small volumes deteriorates
Solution Approach 1:
The patent applies local quality by treating different regions of the volume data with different processing characteristics. Small-volume organs are identified through preliminary processing and given special attention, with the model adjusted to maintain high detection accuracy for these critical regions while efficiently processing the overall dataset.
Solution Approach 2:
The patent implements feedback mechanisms where the machine learning model continuously refines its detection based on the presence and characteristics of organs in each segmented dataset. The model adapts its parameters based on the detected organ volumes, ensuring that small organs are not lost in the segmentation process and maintaining high detection accuracy across all organ sizes.
Data Source
AI summary
There are provided machine learning device and method which can prepare divided data suitable for machine learning from volume data for learning. A machine learning unit (15) calculates detection accuracy of each organ O(j,i) in a predicted mask Pj using a loss function Loss. However, the detection accuracy of the organ O(k,i) with a volume ratio A(k,i)<Th is not calculated. That is, in the predicted mask Pk, the detection accuracy of the organ O(k,i) with a volume ratio that is small to some extent is ignored. The machine learning unit (15) changes each connection load of a neural network (16) from an output layer side to an input layer side according to the loss function Loss.