Portal Dosimetry System Using ML for Discrepancy Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Interpreting portal dosimetry results to distinguish between true positive and false positive differences in radiation therapy is challenging due to the complexity of variables involved, leading to time-consuming and costly analyses for physicists.
Innovation Solution
An expert decision system is provided that uses a database of analyzed portal dosimetry images to perform similarity measurements and classify differences between planning and actual portal images, offering a ranked list of explanations for discrepancies, thereby assisting physicists in determining true positive or false positive differences.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If physicists manually analyze portal dosimetry images to distinguish between true positive and false positive differences, then measurement precision can be maintained, but analysis time and expense increase significantly
Solution Approach 1:
The system creates a virtual copy of the physicist's expertise by training a machine learning model on previously analyzed portal dosimetry images. The model learns to distinguish between true positive and false positive differences by studying annotated examples, then applies this knowledge to new images automatically, eliminating the need for manual review while maintaining diagnostic accuracy.
Solution Approach 2:
The patent replaces the mechanical process of manual visual inspection and expert judgment with an automated machine learning system. The ML model processes portal dosimetry images, compares planned versus actual radiation delivery, and classifies differences automatically, substituting human cognitive processing with computational algorithms that operate continuously without fatigue.
2Reliability
If physicists manually review each portal dosimetry image, then reliability of difference identification is maintained, but productivity decreases due to time-consuming analyses
Solution Approach 1:
The system replicates the reliable judgment of experienced physicists by training ML models on extensively annotated datasets. The models learn the subtle patterns and contextual cues that enable reliable distinction between true positive and false positive differences, achieving expert-level reliability through data-driven learning rather than human review.
Solution Approach 2:
The machine learning system operates continuously without interruption, processing portal dosimetry images as they become available. Unlike manual analysis that requires breaks, attention management, and sequential review, the automated system maintains constant operational capacity, significantly increasing throughput while maintaining consistent diagnostic reliability.
3Measurement precision
If detailed manual analysis is performed to characterize differences, then measurement precision improves, but device complexity increases due to multiple variables involved
Solution Approach 1:
The patent replaces complex manual evaluation procedures with automated machine learning algorithms that internally process multiple variables through learned patterns. The system handles the complexity of distinguishing between true positive and false positive differences by using neural networks that automatically weigh and integrate multiple imaging parameters, anatomical contexts, and treatment plan information without requiring explicit programming of all possible scenarios.
Data Source
Figure 1
Figure 2
Figure 3~4
AI summary
A system for evaluating treatment parameters for a patient undergoing radiotherapy is provided. The system determines a portal dosimetry image (54, 56) showing a difference between a planning image and a portal image (50) both images corresponding to a target region of the patient. The planning image is obtained prior to a radiotherapy treatment session. The portal image (50) is obtained during the radiotherapy treatment session. The system performs a similarity measurement between the portal dosimetry image and prior portal dosimetry images stored in the system. Based on the similarity measurement and the assessments of the prior portal dosimetry images, the system is able to determine when radiation is not being delivered as planned during the radiotherapy treatment session. The system interrupts the radiotherapy treatment session if such determination is made.