Development and Evaluation of Deep Learning-Based Diagnostic Framework for Accurate Differentiation Between Benign and Malignant Breast Tumors Using Histopathological Imaging Data

Abstract

Breast cancer is the most prevalent form of cancer affecting women, characterized by abnormal cell division in breast tissue. Not all tumors pose a significant threat to life, as they can be either benign or malignant. Determining the nature of a tumor—whether benign or malignant—is necessary for guiding appropriate treatment strategies and ensuring patient well-being. In this scenario, medical imaging is becoming a key application of artificial intelligence in healthcare to improve diagnostic accuracy. This study presents the development and evaluation of a deep learning-based tool designed to distinguish between benign and malignant breast tumors using histopathological images. The research used the Breast Cancer Histopathological Image Classification (BreakHis) dataset, which contains 7,909 images from 82 patients across four magnification levels (40X, 100X, 200X, and 400X). The goal is to enhance the accuracy and scalability of breast cancer diagnosis through the application of computer vision and deep learning models. Data preprocessing in this study involved resizing images to a uniform size and applying data augmentation techniques, including random brightness adjustments, flips, and rotations. These methods were employed to improve the models’ ability to handle variations in image orientation and lighting. The study evaluated several deep learning models, including a Convolutional Neural Network (CNN) and 4 transfer learning models including MobileNetV3, EfficientNetB1, VGG16, and ResNet50V2. The findings showed that EfficientNetB1 achieved the highest performance, with a ROC-AUC score of 0.8767, demonstrating strong potential for distinguishing between benign and malignant cases. However, the model also produced a relatively high number of false positives, which is a concern for clinical application. The CNN, although simpler, achieved the highest accuracy, suggesting its potential for use in resource-limited settings. The findings indicate that deep learning models can be applied in breast cancer diagnosis. Further refinement is, however, needed to reduce false positives and ensure the models' reliability in clinical practice.