Brain Metastasis Detection Using Morphology Filtering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for detecting anatomical regions of interest in medical images for radiotherapy treatment planning are time-consuming and lack effective classification capabilities, often relying on high-quality training data that are scarce and requiring manual observation by physicians.
Innovation Solution
A system and method for detecting anatomical regions of interest in medical images using a target detection system that includes a visibility filter, object identifier, morphology filter, and object classifier to automatically classify, position, and segment targets like brain metastases based on morphological features, enhancing visibility and reducing noise, and providing a confidence measure for classification.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Illumination intensity
If pure image processing methods are used to enhance visibility of medical images, then visibility is improved, but classification ability to determine whether an object is a target is lost
Solution Approach 1:
The patent combines pure image processing methods (visibility enhancement) with statistical model-based classification methods. The system applies visibility filters to enhance image quality while simultaneously using trained statistical models to classify objects as targets or non-targets, thereby achieving both improved visibility and retained classification ability
2Adaptability or versatility
If statistical model-based methods are used to identify targets, then classification ability is improved, but dependence on high-quality training data increases which are in short supply
Solution Approach 1:
The system uses the medical images themselves and their inherent morphological features to perform classification, reducing dependence on external training data. The statistical models are trained on available data and then apply morphological analysis to new images, allowing the system to serve itself with minimal external training resources
3Measurement precision
If manual observation by physicians is used to determine targets, then accuracy is improved, but time consumption increases significantly
Solution Approach 1:
The patent introduces a computer-aided detection system as an intermediary between the medical images and the physician. The system automatically performs target detection using statistical models and morphological features, providing assistance to physicians rather than replacing them, thereby reducing time consumption while maintaining accuracy through the intermediary's automated analysis
4Productivity
If computer-aided image classification techniques are used, then time extraction of information is reduced, but effectiveness depends on quality of training data which is in short supply
Solution Approach 1:
The system performs preliminary morphological analysis and feature extraction from the medical images before final classification. By pre-processing the images to enhance visibility and extract relevant morphological features in advance, the system reduces the amount and quality requirements of training data needed for the final classification step, thereby maintaining high productivity while reducing training data dependency
Data Source
Figure 1
Figure 2A
Figure 2B
AI summary
The present disclosure provides systems, methods, and computer- readable storage media for detecting an anatomical region of interest for radiotherapy planning. Embodiments of the present disclosure may identify (810) a plurality of objects in a medical image and select (820) a subset of the objects by applying a morphology filter (516) to the plurality of objects. The morphology filter may determine a morphological feature associated with each of the plurality of objects and exclude at least one object from the subset when the morphological feature of the at least one object exceeds a predetermined threshold. Embodiments of the present disclosure may also classify (830) the objects in the subset into one of a predetermined set of shapes and detect (840) the anatomical region of interest based on the classified objects in the subset.