Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

ContrastDiagnosis: Enhancing Interpretability in Lung Nodule Diagnosis Using Contrastive Learning

Chenglong Wang, Yinqiao Yi, Yida Wang, Chengxiu Zhang, Yun Liu, Kensaku Mori, Mei Yuan, Guang Yang

TL;DR

ContrastDiagnosis incorporates a contrastive learning mechanism to provide a case-based reasoning diagnostic rationale, enhancing the model's transparency and also offers post-hoc interpretability by highlighting similar areas.

Abstract

With the ongoing development of deep learning, an increasing number of AI models have surpassed the performance levels of human clinical practitioners. However, the prevalence of AI diagnostic products in actual clinical practice remains significantly lower than desired. One crucial reason for this gap is the so-called `black box' nature of AI models. Clinicians' distrust of black box models has directly hindered the clinical deployment of AI products. To address this challenge, we propose ContrastDiagnosis, a straightforward yet effective interpretable diagnosis framework. This framework is designed to introduce inherent transparency and provide extensive post-hoc explainability for deep learning model, making them more suitable for clinical medical diagnosis. ContrastDiagnosis incorporates a contrastive learning mechanism to provide a case-based reasoning diagnostic rationale, enhancing the model's transparency and also offers post-hoc interpretability by highlighting similar areas. High diagnostic accuracy was achieved with AUC of 0.977 while maintain a high transparency and explainability.

ContrastDiagnosis: Enhancing Interpretability in Lung Nodule Diagnosis Using Contrastive Learning

TL;DR

ContrastDiagnosis incorporates a contrastive learning mechanism to provide a case-based reasoning diagnostic rationale, enhancing the model's transparency and also offers post-hoc interpretability by highlighting similar areas.

Abstract

With the ongoing development of deep learning, an increasing number of AI models have surpassed the performance levels of human clinical practitioners. However, the prevalence of AI diagnostic products in actual clinical practice remains significantly lower than desired. One crucial reason for this gap is the so-called `black box' nature of AI models. Clinicians' distrust of black box models has directly hindered the clinical deployment of AI products. To address this challenge, we propose ContrastDiagnosis, a straightforward yet effective interpretable diagnosis framework. This framework is designed to introduce inherent transparency and provide extensive post-hoc explainability for deep learning model, making them more suitable for clinical medical diagnosis. ContrastDiagnosis incorporates a contrastive learning mechanism to provide a case-based reasoning diagnostic rationale, enhancing the model's transparency and also offers post-hoc interpretability by highlighting similar areas. High diagnostic accuracy was achieved with AUC of 0.977 while maintain a high transparency and explainability.
Paper Structure (10 sections, 6 equations, 3 figures, 1 table)

This paper contains 10 sections, 6 equations, 3 figures, 1 table.

Table of Contents

  1. Introduction
  2. Related works
  3. Methods
  4. Training
  5. Inference
  6. Explainability
  7. Contrastive Visualization
  8. Confidence Score
  9. Experiments and Results
  10. Discussions and Conclusion

Figures (3)

  • Figure 1: The overview of ContrastDiagnosis network architecture. Aug denotes online data augmentation. 'AlignDist' module is to align two latent representation code and meaure their distance.
  • Figure 2: Several prediction results. The left column shows the query data, while the right-side columns represent the top six most similar cases from support set. The similarity indicators provide a measure of confidence for each prediction.
  • Figure 3: Contrastive regions in query-support pairs. Three cases, aligned with Fig. \ref{['fig:pairs']}, are shown. Each column presents a query-support pair with the best slice of 3D patch highlighted, which may appear slightly different from each pair. The region marked with yellow line is the paired activation region identified by Local-CAM.