Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Clinically inspired enhance Explainability and Interpretability of an AI-Tool for BCC diagnosis based on expert annotation

Iván Matas, Carmen Serrano, Francisca Silva, Amalia Serrano, Tomás Toledo-Pastrana, Begoña Acha

TL;DR

This work tackles the need for interpretable AI in teledermatology for Basal Cell Carcinoma (BCC) diagnosis by coupling a binary BCC/non-BCC classifier with clinically grounded explanations. It introduces a dual XAI framework: a Clinically-inspired XAI that leverages expert-derived pattern labels (via an EM-inferred Standard Reference) and a Clinically-inspired Visual XAI that overlays Grad-CAM maps on dermatologist-segmented patterns. Using a MobileNet-V2 backbone with a three-step training protocol and class-imbalance handling, the approach achieves around 0.90 binary accuracy and 99% detection of BCC when any pattern is present, with Grad-CAM activations aligning closely to clinician regions. The combination of pattern-level explanations and region-focused visual explanations aims to build trust and support early referral workflows in teledermatology, with potential deployment in Andalusia given the model’s lightweight design. Overall, the study demonstrates that clinically aligned explanations can enhance both accuracy and interpretability in AI-assisted skin cancer screening.

Abstract

An AI tool has been developed to provide interpretable support for the diagnosis of BCC via teledermatology, thus speeding up referrals and optimizing resource utilization. The interpretability is provided in two ways: on the one hand, the main BCC dermoscopic patterns are found in the image to justify the BCC/Non BCC classification. Secondly, based on the common visual XAI Grad-CAM, a clinically inspired visual explanation is developed where the relevant features for diagnosis are located. Since there is no established ground truth for BCC dermoscopic features, a standard reference is inferred from the diagnosis of four dermatologists using an Expectation Maximization (EM) based algorithm. The results demonstrate significant improvements in classification accuracy and interpretability, positioning this approach as a valuable tool for early BCC detection and referral to dermatologists. The BCC/non-BCC classification achieved an accuracy rate of 90%. For Clinically-inspired XAI results, the detection of BCC patterns useful to clinicians reaches 99% accuracy. As for the Clinically-inspired Visual XAI results, the mean of the Grad-CAM normalized value within the manually segmented clinical features is 0.57, while outside this region it is 0.16. This indicates that the model struggles to accurately identify the regions of the BCC patterns. These results prove the ability of the AI tool to provide a useful explanation.

Clinically inspired enhance Explainability and Interpretability of an AI-Tool for BCC diagnosis based on expert annotation

TL;DR

This work tackles the need for interpretable AI in teledermatology for Basal Cell Carcinoma (BCC) diagnosis by coupling a binary BCC/non-BCC classifier with clinically grounded explanations. It introduces a dual XAI framework: a Clinically-inspired XAI that leverages expert-derived pattern labels (via an EM-inferred Standard Reference) and a Clinically-inspired Visual XAI that overlays Grad-CAM maps on dermatologist-segmented patterns. Using a MobileNet-V2 backbone with a three-step training protocol and class-imbalance handling, the approach achieves around 0.90 binary accuracy and 99% detection of BCC when any pattern is present, with Grad-CAM activations aligning closely to clinician regions. The combination of pattern-level explanations and region-focused visual explanations aims to build trust and support early referral workflows in teledermatology, with potential deployment in Andalusia given the model’s lightweight design. Overall, the study demonstrates that clinically aligned explanations can enhance both accuracy and interpretability in AI-assisted skin cancer screening.

Abstract

An AI tool has been developed to provide interpretable support for the diagnosis of BCC via teledermatology, thus speeding up referrals and optimizing resource utilization. The interpretability is provided in two ways: on the one hand, the main BCC dermoscopic patterns are found in the image to justify the BCC/Non BCC classification. Secondly, based on the common visual XAI Grad-CAM, a clinically inspired visual explanation is developed where the relevant features for diagnosis are located. Since there is no established ground truth for BCC dermoscopic features, a standard reference is inferred from the diagnosis of four dermatologists using an Expectation Maximization (EM) based algorithm. The results demonstrate significant improvements in classification accuracy and interpretability, positioning this approach as a valuable tool for early BCC detection and referral to dermatologists. The BCC/non-BCC classification achieved an accuracy rate of 90%. For Clinically-inspired XAI results, the detection of BCC patterns useful to clinicians reaches 99% accuracy. As for the Clinically-inspired Visual XAI results, the mean of the Grad-CAM normalized value within the manually segmented clinical features is 0.57, while outside this region it is 0.16. This indicates that the model struggles to accurately identify the regions of the BCC patterns. These results prove the ability of the AI tool to provide a useful explanation.
Paper Structure (13 sections, 4 figures, 4 tables)

This paper contains 13 sections, 4 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Methodology
  3. Database
  4. Label codification
  5. Standard Reference (SR) Inferring
  6. Clinically-inspired XAI: BCC diagnosis with additional label information
  7. Clinically-inspired Visual XAI
  8. Results
  9. Implementation details
  10. Clinically-inspired XAI: BCC diagnosis with additional label information
  11. Clinically-inspired Visual XAI
  12. Discussion
  13. Acknowledgements

Figures (4)

  • Figure 1: Illustration showing an example of the current workflow of BCC diagnosis by teledermatology in orange and the proposed workflow in green.
  • Figure 2: BCC dermoscopic criteria and Pigment Network as a negative criterion.
  • Figure 3: Example of the specialist segmentation job.
  • Figure 4: Demonstration of various interpretative visualizations used in skin lesion analysis, combining expert segmentation and Grad-CAM heatmaps to aid in understanding the model's decision-making process. This example is from a correct prediction sample.