Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Training-Aware Risk Control for Intensity Modulated Radiation Therapies Quality Assurance with Conformal Prediction

Kevin He, David Adam, Sarah Han-Oh, Anqi Liu

TL;DR

This work tackles the high-cost IMRT QA bottleneck by introducing a training-aware conformal risk control framework that integrates clinical gamma passing rate thresholds into ML training to triage treatment plans. It merges conformal prediction with risk-controlled decision-making, yielding calibrated prediction bands while emphasizing a conservative lower-bound decision to maximize safety. Across two real-world JHH IMRT plan datasets, the approach achieves high sensitivity and superior specificity with substantial reductions in measurement, even under distribution shifts, demonstrating practical potential to streamline IMRT QA workflows. The study highlights the applicability of distribution-free uncertainty methods in a safety-critical clinical setting and outlines avenues for deployment, calibration, and future improvements under changing data conditions.

Abstract

Measurement quality assurance (QA) practices play a key role in the safe use of Intensity Modulated Radiation Therapies (IMRT) for cancer treatment. These practices have reduced measurement-based IMRT QA failure below 1%. However, these practices are time and labor intensive which can lead to delays in patient care. In this study, we examine how conformal prediction methodologies can be used to robustly triage plans. We propose a new training-aware conformal risk control method by combining the benefit of conformal risk control and conformal training. We incorporate the decision making thresholds based on the gamma passing rate, along with the risk functions used in clinical evaluation, into the design of the risk control framework. Our method achieves high sensitivity and specificity and significantly reduces the number of plans needing measurement without generating a huge confidence interval. Our results demonstrate the validity and applicability of conformal prediction methods for improving efficiency and reducing the workload of the IMRT QA process.

Training-Aware Risk Control for Intensity Modulated Radiation Therapies Quality Assurance with Conformal Prediction

TL;DR

This work tackles the high-cost IMRT QA bottleneck by introducing a training-aware conformal risk control framework that integrates clinical gamma passing rate thresholds into ML training to triage treatment plans. It merges conformal prediction with risk-controlled decision-making, yielding calibrated prediction bands while emphasizing a conservative lower-bound decision to maximize safety. Across two real-world JHH IMRT plan datasets, the approach achieves high sensitivity and superior specificity with substantial reductions in measurement, even under distribution shifts, demonstrating practical potential to streamline IMRT QA workflows. The study highlights the applicability of distribution-free uncertainty methods in a safety-critical clinical setting and outlines avenues for deployment, calibration, and future improvements under changing data conditions.

Abstract

Measurement quality assurance (QA) practices play a key role in the safe use of Intensity Modulated Radiation Therapies (IMRT) for cancer treatment. These practices have reduced measurement-based IMRT QA failure below 1%. However, these practices are time and labor intensive which can lead to delays in patient care. In this study, we examine how conformal prediction methodologies can be used to robustly triage plans. We propose a new training-aware conformal risk control method by combining the benefit of conformal risk control and conformal training. We incorporate the decision making thresholds based on the gamma passing rate, along with the risk functions used in clinical evaluation, into the design of the risk control framework. Our method achieves high sensitivity and specificity and significantly reduces the number of plans needing measurement without generating a huge confidence interval. Our results demonstrate the validity and applicability of conformal prediction methods for improving efficiency and reducing the workload of the IMRT QA process.
Paper Structure (25 sections, 1 theorem, 22 equations, 2 figures, 8 tables)

This paper contains 25 sections, 1 theorem, 22 equations, 2 figures, 8 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. IMRT QA Prediction with Machine Learning Models
  4. Conformal Prediction in the Healthcare Domain
  5. Methodology
  6. Conformal Prediction
  7. Conformal Quantile Regression
  8. Conformal Risk Control
  9. Conformal Training
  10. Training-Aware Conformal Risk Control for IMRT QA
  11. Experiments
  12. Datasets
  13. Experimental Setup
  14. Evaluation Metrics
  15. Data Preprocessing
  16. ...and 10 more sections

Key Result

Theorem 1

Assume that the loss function is defined as: and that $\ell(C_\lambda(X_i), Y_i)$ is non-increasing in $\lambda$, right-continuous, and Then

Figures (2)

  • Figure 1: The workflow of a radiation oncology department for delivering IMRT treatment. The IMRT process involves multiple different disciplines and professionals in the workflow to design and implement a personalized treatment plan based on a patient's unique disease condition. The IMRT QA process is the evaluation process after a plan is designed and before a plan is deployed in the treatment. It is a safety-critical task as we do not want to deliver low-quality treatment to patients. Overdosing healthy organs can result in radiation-induced side effects, while underdosing the tumor reduces the tumor control probability, thereby decreasing the efficacy of radiotherapy.
  • Figure 2: Depiction of coronal absorbed dose distribution in the A) patient and correspondingly in the B) quality assurance phantom detector plane.

Theorems & Definitions (1)

  • Theorem 1