Patchwise CNN Rib Fracture Detection With Activation Maps
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Diagnosis of rib fractures from chest medical images is often ambiguous and prone to misdiagnosis due to the complexity of the rib structure and the reliance on human expertise, leading to potential overlooking of fractures in critical areas like the heart and lungs.
Innovation Solution
A method and device using a convolutional neural network (CNN) model to analyze chest medical images, generating a class activation map (CAM) that visually indicates rib fracture points and probabilities, enabling quick and accurate diagnosis.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If medical staff directly examine chest medical images to determine rib fractures, then diagnostic experience and expertise are utilized, but the diagnosis may be ambiguous and fractures may be overlooked due to human limitations and time consumption
Solution Approach 1:
A CNN-based AI system serves as an intermediary between the medical image and the medical staff. The system processes chest medical images and generates class activation maps that highlight potential rib fracture locations, providing objective diagnostic assistance that complements medical expertise while reducing subjective variability and time consumption
Solution Approach 2:
The patent replaces the mechanical process of human visual examination with an automated AI-based image analysis system. The CNN model automatically detects rib fractures by learning from training data, substituting the manual examination process with an automated computational system that provides consistent and rapid diagnostic support
2Measurement precision
If additional special tests such as computed tomography and magnetic resonance imaging are performed to confirm fracture patterns, then diagnostic accuracy is improved, but examination complexity and time consumption increase
Solution Approach 1:
The patent segments the diagnostic process into two stages: initial screening using the CNN-based AI system on standard chest medical images, and follow-up detailed examination only for cases where fractures are detected or suspected. This segmentation allows most cases to be handled with simple imaging while reserving complex imaging for specific needs
Solution Approach 2:
The AI system performs a preliminary analysis that covers the entire chest region, providing partial diagnostic information that can confirm or rule out fractures in many cases. This partial action (initial AI screening) reduces the need for full additional imaging in cases where the AI provides sufficient diagnostic confidence
3Reliability
If the entire chest medical image is analyzed at once to detect rib fractures, then comprehensive coverage is achieved, but processing time and computational load increase
Solution Approach 1:
The patent divides the chest medical image into multiple patches and processes them independently through the CNN model. This segmentation allows parallel processing of image regions, reducing computational load and processing time while maintaining comprehensive coverage through subsequent reconstruction of the class activation map
Solution Approach 2:
The patent transforms the problem from analyzing the entire 2D image at once to processing multiple smaller 2D patches and then reconstructing the results in a new dimensional framework (the class activation map). This dimensional transformation enables efficient parallel processing while preserving comprehensive diagnostic information
Data Source
Figure 1A
Figure 1B
Figure 1C
AI summary
Provided are a method of providing information on a rib fracture, and a device for providing information on a rib fracture by using same. The method of providing information on a rib fracture, implemented by a processor, comprises the steps of: receiving a chest medical image of a subject, extracting the chest medical image patch by patch, determining by using a convolutional neutral network model pretrained to determine rib fracture probability by using a patch as an input, rib fracture probability by using the patch as an input; generating a class activation map (CAM) patch by patch, and reconstructing the entire class activation map generated patch by patch, to correspond to the entire chest medical image.