CT Nodule State Change Prediction for pGGO Follow-Up Timing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current CADx devices for lung cancer diagnosis struggle to provide timely information on the potential solid transformation of pure ground-glass opacity (pGGO) nodules, which are associated with uncertainty in management pathways due to their unpredictable behavior, leading to prolonged and potentially unnecessary follow-up periods.
Innovation Solution
A CADx system with a feature state change prediction circuit that analyzes medical images, particularly CT scans, to predict the time frame for transformation from pGGO to semi-solid or solid nodules, incorporating machine learning models to determine nodule attenuation and activate specific prediction tasks based on clinical parameters.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If CADx devices provide only malignancy risk scores, then the device complexity is low and ease of operation is high, but the loss of information occurs as urgency indication is not provided
Solution Approach 1:
The system segments the risk assessment into two distinct components: malignancy risk score and solid transformation probability. This segmentation allows the device to provide comprehensive information (including urgency indication) while maintaining clear, separate output modules that do not significantly increase operational complexity.
Solution Approach 2:
The system adds a new dimension to the risk assessment by introducing temporal prediction (solid transformation probability within specific time frames). This transforms the static malignancy risk score into a dynamic assessment that includes both malignancy likelihood and transformation urgency, providing comprehensive information without requiring complete system redesign.
2Reliability
If CADx devices provide only malignancy risk scores, then the device complexity is low, but the reliability is insufficient for clinical decision-making
Solution Approach 1:
The system performs preliminary classification of GGO nodules based on their likely transformation behavior. By pre-categorizing nodules into time-frame groups (e.g., likely to transform within 6 months, 1-2 years, or longer), the system provides reliable guidance for clinical decision-making while using established machine learning models that do not significantly increase device complexity.
Solution Approach 2:
The system introduces an intermediary prediction layer that bridges malignancy risk assessment and clinical urgency determination. The solid transformation probability acts as a mediator that translates complex imaging features into clinically actionable urgency indicators, enhancing reliability without requiring direct complex clinical decision logic in the device.
3Productivity
If prolonged follow-ups are required for GGO nodules, then the reliability of diagnosis is maintained, but the loss of time occurs and productivity is reduced
Solution Approach 1:
The system dynamically adjusts follow-up recommendations based on predicted solid transformation probability. Instead of uniform prolonged follow-ups for all GGO nodules, the system adapts follow-up intervals to individual nodule characteristics, enabling earlier intervention for high-risk cases and reducing unnecessary follow-ups for low-risk cases, thereby improving both productivity and maintaining diagnosis reliability.
Solution Approach 2:
The system changes the key parameter from static malignancy risk to dynamic transformation probability over time. By predicting when solid transformation is likely to occur, the system optimizes follow-up timing parameters, allowing clinicians to intervene at the optimal moment rather than relying on fixed prolonged follow-up schedules, thus improving productivity while maintaining reliability.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A CADx system for analysing medical images to monitor at least one feature on the image to predict at least one of a change of the state or maintenance of the current state of a monitored feature within a time frame is described. The system comprising: an input circuit for receiving at least one medical input image; a feature state change circuit for analysing the received input image and predicting a state change comprising: a state change predictor to predict a state change of the monitored feature within the time frame; and an output circuit to output an indication of the change of state or maintenance of the current state of the monitored feature within the time frame based on the prediction of the feature state change predictor. A method of training a feature state change prediction circuit using Machine Learning is also described.