Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Unsupervised Out-of-Distribution Detection in Medical Imaging Using Multi-Exit Class Activation Maps and Feature Masking

Yu-Jen Chen, Xueyang Li, Yiyu Shi, Tsung-Yi Ho

TL;DR

This work tackles unreliable OOD detection in medical imaging by introducing MECAM, which fuses CAMs from multiple network exits and applies inverted CAM masking to induce a feature-space shift between in-distribution and out-of-distribution inputs. The OOD score is computed as $\text{Score}_{OOD} = \frac{1}{d} \sum_{i=1}^{d} (v_i - v'_i)^2$ based on embeddings before and after masking, with CAMs generated at different depths and weighted by exit confidences. MECAM is evaluated on ISIC19 and PathMNIST as ID data and RSNA Pneumonia, COVID-19, HeadCT, and iSUN as OOD data, using AUC and FPR95 as metrics, and it consistently outperforms state-of-the-art methods. The approach enhances reliability and interpretability of medical imaging models and extends CAM-based OOD detection to diverse clinical scenarios.

Abstract

Out-of-distribution (OOD) detection is essential for ensuring the reliability of deep learning models in medical imaging applications. This work is motivated by the observation that class activation maps (CAMs) for in-distribution (ID) data typically emphasize regions that are highly relevant to the model's predictions, whereas OOD data often lacks such focused activations. By masking input images with inverted CAMs, the feature representations of ID data undergo more substantial changes compared to those of OOD data, offering a robust criterion for differentiation. In this paper, we introduce a novel unsupervised OOD detection framework, Multi-Exit Class Activation Map (MECAM), which leverages multi-exit CAMs and feature masking. By utilizing mult-exit networks that combine CAMs from varying resolutions and depths, our method captures both global and local feature representations, thereby enhancing the robustness of OOD detection. We evaluate MECAM on multiple ID datasets, including ISIC19 and PathMNIST, and test its performance against three medical OOD datasets, RSNA Pneumonia, COVID-19, and HeadCT, and one natural image OOD dataset, iSUN. Comprehensive comparisons with state-of-the-art OOD detection methods validate the effectiveness of our approach. Our findings emphasize the potential of multi-exit networks and feature masking for advancing unsupervised OOD detection in medical imaging, paving the way for more reliable and interpretable models in clinical practice.

Unsupervised Out-of-Distribution Detection in Medical Imaging Using Multi-Exit Class Activation Maps and Feature Masking

TL;DR

This work tackles unreliable OOD detection in medical imaging by introducing MECAM, which fuses CAMs from multiple network exits and applies inverted CAM masking to induce a feature-space shift between in-distribution and out-of-distribution inputs. The OOD score is computed as based on embeddings before and after masking, with CAMs generated at different depths and weighted by exit confidences. MECAM is evaluated on ISIC19 and PathMNIST as ID data and RSNA Pneumonia, COVID-19, HeadCT, and iSUN as OOD data, using AUC and FPR95 as metrics, and it consistently outperforms state-of-the-art methods. The approach enhances reliability and interpretability of medical imaging models and extends CAM-based OOD detection to diverse clinical scenarios.

Abstract

Out-of-distribution (OOD) detection is essential for ensuring the reliability of deep learning models in medical imaging applications. This work is motivated by the observation that class activation maps (CAMs) for in-distribution (ID) data typically emphasize regions that are highly relevant to the model's predictions, whereas OOD data often lacks such focused activations. By masking input images with inverted CAMs, the feature representations of ID data undergo more substantial changes compared to those of OOD data, offering a robust criterion for differentiation. In this paper, we introduce a novel unsupervised OOD detection framework, Multi-Exit Class Activation Map (MECAM), which leverages multi-exit CAMs and feature masking. By utilizing mult-exit networks that combine CAMs from varying resolutions and depths, our method captures both global and local feature representations, thereby enhancing the robustness of OOD detection. We evaluate MECAM on multiple ID datasets, including ISIC19 and PathMNIST, and test its performance against three medical OOD datasets, RSNA Pneumonia, COVID-19, and HeadCT, and one natural image OOD dataset, iSUN. Comprehensive comparisons with state-of-the-art OOD detection methods validate the effectiveness of our approach. Our findings emphasize the potential of multi-exit networks and feature masking for advancing unsupervised OOD detection in medical imaging, paving the way for more reliable and interpretable models in clinical practice.
Paper Structure (8 sections, 3 equations, 2 figures, 3 tables)

This paper contains 8 sections, 3 equations, 2 figures, 3 tables.

Table of Contents

  1. Introduction
  2. MECAM-OODD
  3. Experiments
  4. Results
  5. Comparison with State-of-the-Art
  6. Comparison with Different CAM Approaches
  7. Ablation Study
  8. Conclusion

Figures (2)

  • Figure 1: Illustration showing that masking the image with the CAM produces significant changes in feature representations. (a) Example of ID and OOD images with their corresponding masked images. (b) Visualization of the features of images and masked images for both ID and OOD data. ID: ISIC dataset, OOD: RSNA Pneumonia dataset.
  • Figure 2: An illustration of the proposed MECAM framework for OOD detection.