Scholarlink

In this article, Explainable Cubic Attention-based Autoencoder (ECAbA) model is proposed for skin cancer classification. The proposed model has a data augmentation module, which transform an imbalanced HAM10000 dataset into a refined and balanced form. In the second step, Cubic Attention-based Autoencoder (CAbA) model is proposed, which has 5 3D convolutional layers, 2 attention modules and one flatten, dense and neural successive cancellation layers. The trained model is then used for three explainable models, i.e., LIME, Grad-CAM, and Kernel SHAP, to make the explanations of predicted images. The outputs of these explainer modules make the predictions of CAbA model explainable and reliable. For fostering confidence among medical practitioners, the proposed model can assist in the adoption of AI approaches.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Explainable Cubic Attention-Based Autoencoder for Skin Cancer Classification

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Inzamam Mashood Nasir,
  • Sara Tehsin,
  • Robertas Damaševičius,
  • Adam Zielonka,
  • Marcin Woźniak

摘要

In this article, Explainable Cubic Attention-based Autoencoder (ECAbA) model is proposed for skin cancer classification. The proposed model has a data augmentation module, which transform an imbalanced HAM10000 dataset into a refined and balanced form. In the second step, Cubic Attention-based Autoencoder (CAbA) model is proposed, which has 5 3D convolutional layers, 2 attention modules and one flatten, dense and neural successive cancellation layers. The trained model is then used for three explainable models, i.e., LIME, Grad-CAM, and Kernel SHAP, to make the explanations of predicted images. The outputs of these explainer modules make the predictions of CAbA model explainable and reliable. For fostering confidence among medical practitioners, the proposed model can assist in the adoption of AI approaches.