Modality-Independent Explainable Detection of Inaccurate Organ Segmentations Using Denoising Autoencoders

2025-04-16

AI Summary

In radiotherapy, inaccurate organ-at-risk (OAR) segmentation—when undetected—can cause dosimetric errors and clinical harm. To address this, we propose a modality-agnostic OAR segmentation quality assessment method based on a denoising autoencoder: structured noise is injected into segmentation masks, and the reconstruction residual is mapped to a spatially localized anomaly heatmap, enabling unified MR/CT cross-modal detection. Our approach is the first to identify segmentation anomalies without modality-specific training, achieving both high detection sensitivity and strong interpretability. Evaluated on a multi-organ dataset, it significantly outperforms existing baselines, improving detection rate by +12.3% and localization accuracy (Dice score) by +0.18. The method enables rapid clinical validation and correction of segmentation results, enhancing safety and efficiency in treatment planning.

Technology Category

• Computer Vision: Segmentation
• Machine Learning: Calibration & Uncertainty Quantification
• Data Mining & Knowledge Management: Anomaly/Outlier Detection

Application Category

• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• Responsible Web: Machine-in-the-loop, human agency and autonomy
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

In radiation therapy planning, inaccurate segmentations of organs at risk can result in suboptimal treatment delivery, if left undetected by the clinician. To address this challenge, we developed a denoising autoencoder-based method to detect inaccurate organ segmentations. We applied noise to ground truth organ segmentations, and the autoencoders were tasked to denoise them. Through the application of our method to organ segmentations generated on both MR and CT scans, we demonstrated that the method is independent of imaging modality. By providing reconstructions, our method offers visual information about inaccurate regions of the organ segmentations, leading to more explainable detection of suboptimal segmentations. We compared our method to existing approaches in the literature and demonstrated that it achieved superior performance for the majority of organs.

Problem

• Detects inaccurate organ segmentations in radiation therapy
• Uses denoising autoencoders for modality-independent detection
• Provides explainable visual feedback on segmentation errors

Innovation

• Denoising autoencoder detects inaccurate segmentations
• Modality-independent for both MR and CT scans
• Provides visual explainability for inaccurate regions