Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Evidential Concept Embedding Models: Towards Reliable Concept Explanations for Skin Disease Diagnosis

Yibo Gao, Zheyao Gao, Xin Gao, Yuanye Liu, Bomin Wang, Xiahai Zhuang

TL;DR

An evidential Concept Embedding Model (evi-CEM), which employs evidential learning to model the concept uncertainty and introduces concept uncertainty for effective test-time intervention to enhance concept explanations' reliability for both supervised and label-efficient settings.

Abstract

Due to the high stakes in medical decision-making, there is a compelling demand for interpretable deep learning methods in medical image analysis. Concept Bottleneck Models (CBM) have emerged as an active interpretable framework incorporating human-interpretable concepts into decision-making. However, their concept predictions may lack reliability when applied to clinical diagnosis, impeding concept explanations' quality. To address this, we propose an evidential Concept Embedding Model (evi-CEM), which employs evidential learning to model the concept uncertainty. Additionally, we offer to leverage the concept uncertainty to rectify concept misalignments that arise when training CBMs using vision-language models without complete concept supervision. With the proposed methods, we can enhance concept explanations' reliability for both supervised and label-efficient settings. Furthermore, we introduce concept uncertainty for effective test-time intervention. Our evaluation demonstrates that evi-CEM achieves superior performance in terms of concept prediction, and the proposed concept rectification effectively mitigates concept misalignments for label-efficient training. Our code is available at https://github.com/obiyoag/evi-CEM.

Evidential Concept Embedding Models: Towards Reliable Concept Explanations for Skin Disease Diagnosis

TL;DR

An evidential Concept Embedding Model (evi-CEM), which employs evidential learning to model the concept uncertainty and introduces concept uncertainty for effective test-time intervention to enhance concept explanations' reliability for both supervised and label-efficient settings.

Abstract

Due to the high stakes in medical decision-making, there is a compelling demand for interpretable deep learning methods in medical image analysis. Concept Bottleneck Models (CBM) have emerged as an active interpretable framework incorporating human-interpretable concepts into decision-making. However, their concept predictions may lack reliability when applied to clinical diagnosis, impeding concept explanations' quality. To address this, we propose an evidential Concept Embedding Model (evi-CEM), which employs evidential learning to model the concept uncertainty. Additionally, we offer to leverage the concept uncertainty to rectify concept misalignments that arise when training CBMs using vision-language models without complete concept supervision. With the proposed methods, we can enhance concept explanations' reliability for both supervised and label-efficient settings. Furthermore, we introduce concept uncertainty for effective test-time intervention. Our evaluation demonstrates that evi-CEM achieves superior performance in terms of concept prediction, and the proposed concept rectification effectively mitigates concept misalignments for label-efficient training. Our code is available at https://github.com/obiyoag/evi-CEM.
Paper Structure (20 sections, 14 equations, 4 figures, 2 tables)

This paper contains 20 sections, 14 equations, 4 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Method
  3. Evidential Concept Embedding Model
  4. Model architecture:
  5. Uncertainty modelling:
  6. Training paradigm:
  7. Concept Rectification for Label-efficient Training
  8. Pretraining ECBL with VLMs:
  9. Learning CAVs of the misaligned concepts:
  10. Incorporating CAVs into evi-CEM training:
  11. Uncertainty-aware Intervention
  12. Experiments
  13. Experimental Setup
  14. Dataset:
  15. Implementation details:
  16. ...and 5 more sections

Figures (4)

  • Figure 1: (a) An illustration of CBMs. (b) Over-confident examples of CEMs. The orange and blue bars are used to visually represent the probabilities supporting and opposing the concepts, respectively.
  • Figure 2: The architecture of Evidential Concept Embedding Model (evi-CEM). The model primarily comprises a backbone network $\Psi(\cdot)$, an evidential concept bottleneck layer (ECBL) and a task predictor $f(\cdot)$.
  • Figure 3: The process of ECBL pretraining in conjunction with VLMs. The image and text encoders of VLMs remain frozen to extract embeddings for concept estimation using \ref{['eq:estimate_c_k']}. The ECBL is trained with the estimated concept labels to minimize variational concept loss.
  • Figure 4: (a) A qualitative result of uncertainty-aware intervention. $\hat{\mathbf{c}}$ and $\mathbf{u}$ denotes the concept prediction and its uncertainty respectively. (b) Test diagnosis AUC of interventions on evi-CEM and their corresponding standard deviations.