An explainable AI-driven deep neural network for accurate breast cancer detection from histopathological and ultrasound images

Sep 23, 2026

Breast cancer detection requires early, accurate diagnosis to improve prognosis and reduce mortality, yet manual image interpretation is complex and variable, and many deep learning models lack both accuracy and interpretability. This study presents the DNBCD explainable AI framework using Densenet121 with custom CNN layers and transfer learning to classify breast cancer from histopathological and ultrasound images. Preprocessing, normalization, resizing, augmentation, and class-weighting handle imbalance. Grad-CAM visualizations provide prediction justifications, and experiments on B-400x and BUSI report strong accuracies of 93.97% and 89.87%.

LanguageEnglish

CategoryHealthcare

FormatPDF

Pages34

Reading Time1 h 26 min

AccessFree to download

Background and motivation

Limitations of traditional diagnosis

Need for explainable and accurate models

Proposed DNBCD approach

Model architecture and transfer learning

Preprocessing, augmentation, and class imbalance handling

Interpretability with Grad-CAM

Experimental setup and results

Datasets and data splitting

Accuracy outcomes on B-400x and BUSI

Conclusion and availability

[www. nature.com/scientificreports](www. nature.com/scientificreports) OPEN An explainable AI-driven deep neural network for accurate breast cancer detection from histopathological and ultrasound images Md. Romzan Alom1, FahmidAl Farid2, Muhammad Aminur Rahaman1􀀍, Anichur Rahman3,4􀀍, Tanoy Debnath5, Abu Saleh Musa Miah6 & Sarina Mansor2􀀍 Breast cancer represents a significant global health challenge, which makes it essential to detect breast cancer early and accurately to improve patient prognosis and reduce mortality rates. However, traditional diagnostic processes relying on manual analysis of medical images are inherently complex and subject to variability between observers, highlighting the urgent need for robust automated breast cancer detection systems. While deep learning has demonstrated potential, many current models struggle with limited accuracy and lack of interpretability. This research introduces the Deep Neural Breast Cancer Detection (DNBCD) model, an explainable AI-based framework that utilizes deep learning methods for classifying breast cancer using histopathological and ultrasound images. The proposed model employs Densenet121 as a foundation, integrating customized Convolutional Neural Network (CNN) layers including GlobalAveragePooling2D, Dense, and Dropout layers along with transfer learning to achieve both high accuracy and interpretability for breast cancer diagnosis. The proposed DNBCD model integrates several preprocessing techniques, including image normalization and resizing, and augmentation techniques to enhance the model’s robustness and address class imbalances using class weight. It employs Grad-CAM (Gradient-weighted Class Activation Mapping) to offer visual justifications for its predictions, increasing trust and transparency among healthcare providers. The model was assessed using two benchmark datasets: Breakhis-400x (B-400x) and Breast Ultrasound Images Dataset (BUSI) containing 1820 and 1578 images, respectively. We systematically divided the datasets into training (70%), testing (20%,) and validation (10%) sets, ensuring efficient model training and evaluation obtaining accuracies of 93.97% for B-400x dataset having benign and malignant classes and 89.87% for BUSI dataset having benign, malignant, and normal classes for breast cancer detection. Experimental results demonstrate that the proposed DNBCD model significantly outperforms existing state-of-the-art approaches with potential uses in clinical environments. We also made all the materials publicly accessible for the research community at: [https:/](https:/)/[github.com/romzanalom/XAI-Based-Deep-Neural-Breast-Cancer-Detection](github.com/romzanalom/XAI-Based-Deep-Neural-Breast-Cancer-Detection). Keywords Breast cancer, XAI, DNBCD, CNN, Transfer learning, Breakhis-400x, BUSI, Grad-CAM Cancer remains a critical public health challenge worldwide, and Breast Cancer (BC) stands as the most common malignancy globally, representing approximately 12.4% of all new cancer cases each year1,2. In the United States alone, 2024 saw an estimated 313,520 new cases of invasive BC and 310,720 new cases of non- 1Department of Computer Science and Engineering, Green University of Bangladesh (GUB), Purbachal American City, Kanchon, Dhaka 1460, Bangladesh. 2Faculty of Artificial Intelligence and Engineering, Multimedia University, 63100 Cyberjaya, Malaysia. 3Department of Computer Science and Engineering, National Institute of Textile Engineering and Research (NITER), Constituent Institute of the University of Dhaka, Savar, Dhaka 1350, Bangladesh. 4Department of Computer Science and Engineering, Mawlana Bhashani Science and Technology University, Tangail, Bangladesh. 5Department of Computer Science, Stony Brook University, Stony Brook, NY, USA. 6Department of Computer Science and Engineering, Bangladesh Army University of Science and Technology (BAUST), Nilphamari, Bangladesh. 􀀍 email: [aminur@cse.green.edu.bd](aminur@cse.green.edu.bd);

FAQs

What problem does DNBCD address in breast cancer detection?

DNBCD targets the need for early, accurate detection while reducing the variability and complexity of manual medical image analysis. It also focuses on improving interpretability compared with many existing deep learning models.

How does DNBCD provide explainability for its predictions?

DNBCD uses Grad-CAM (Gradient-weighted Class Activation Mapping) to generate visual justifications that highlight regions contributing to each prediction.

Which datasets and evaluation split were used to test the model?

The model was evaluated on two benchmark datasets: Breakhis-400x (B-400x) and BUSI. The datasets were divided into training (70%), testing (20%), and validation (10%) sets.

Sep 23, 2026

Uploaded by Aria Callaghan