Autonomous 3D Anatomical Segmentation via CNN Multidimensional Processing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Low-quality medical images from CT and MRI scans, particularly those with low dose settings, pose challenges in identifying anatomical structures during image-guided surgery and diagnostics, leading to decreased accuracy and increased radiation exposure.
Innovation Solution
A method for autonomous multidimensional segmentation of anatomical structures using a convolutional neural network (CNN) that preprocesses low-quality images, performs human-assisted supervised training, and applies data augmentation to enhance image quality, allowing for accurate segmentation of 3D scan volumes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Object-affected harmful factors
If low dose settings are used for CT scans, then radiation exposure is reduced, but image quality deteriorates making it difficult to identify anatomical structures
Solution Approach 1:
The system performs preliminary actions by pre-processing low-quality images through multiple steps including denoising, enhancement, and generation of synthetic high-quality image pairs. This preliminary processing occurs before the actual segmentation task, transforming degraded input images into usable formats that enable accurate anatomical structure identification while maintaining low radiation dose benefits
Solution Approach 2:
The patent introduces an intermediary approach by creating a bridge between low-dose images and high-quality requirements through intermediate processing stages. The system uses intermediate representations and synthesized image pairs that mediate between the limited quality of low-dose inputs and the quality requirements for accurate segmentation, enabling both low radiation exposure and high measurement precision
2Measurement precision
If high dose settings are used for CT scans, then image quality is improved for adequate visualization, but radiation exposure to the patient increases
Solution Approach 1:
The system creates copies of low-dose images through synthetic image generation techniques. By generating synthetic high-quality image pairs from low-dose inputs, the system produces alternative representations that maintain diagnostic quality without requiring high radiation doses, effectively copying the quality characteristics without the harmful radiation exposure
3Device complexity
If machine learning applications are used with low quality image datasets, then processing is simplified, but accuracy and efficacy of navigated tools decrease
Solution Approach 1:
The system performs preliminary image quality improvement actions before feeding data to machine learning models. By pre-processing images to enhance quality and generate synthetic high-quality pairs, the system ensures that the input data to ML algorithms is of sufficient quality, thereby maintaining high accuracy and efficacy while keeping the overall processing pipeline manageable
Solution Approach 2:
The patent implements feedback mechanisms where the system continuously evaluates image quality and adjusts processing parameters accordingly. The synthetic image generation process incorporates feedback loops that compare generated images against ground truth annotations, allowing the system to iteratively improve quality until accurate segmentation results are achieved
Data Source
Figure 1
Figure 2A~2B
Figure 2C~2D-1
AI summary
A method for autonomous multidimensional segmentation of anatomical structures from three-dimensional (3D) scan volumes, the method comprising the following steps: receiving (301) the 3D scan volume comprising a set of medical scan images comprising the anatomical structures; automatically defining (304) succeeding multidimensional regions of input data used for further processing; autonomously processing (306), by means of a pre-trained segmentation convolutional neural network (CNN) (400), the defined multidimensional regions to determine weak segmentation results that define a probable 3D shape, location, and size of the anatomical structures; automatically combining (307) multiple weak segmentation results by determining segmented voxels that overlap on the weak segmentation results, to obtain raw strong segmentation results with improved accuracy of the segmentation; autonomously filtering (308) the raw strong segmentation results with a predefined set of filters and parameters for enhancing shape, location, size and continuity of the anatomical structures to obtain filtered strong segmentation results; and autonomously identifying (309) a plurality of classes of the anatomical structures from the filtered strong segmentation results.