Advanced Deep Learning Approach for Breast Cancer Detection Using Ultrasound Images
G. Ramkumar, Ibrahim Mohammad Khrais, Md. Tabil Ahammed, P. Illavarason, J. Bino, K Swetha
Abstract
Breast cancer often remains undetected in many women until it reaches an advanced stage, posing a significant global public health concern. Early detection is crucial for ensuring effective treatment and improving patient outcomes. While several studies have employed pre-trained deep learning models for breast tumor identification, these models typically involve numerous layers and parameters, demanding substantial computational resources. To address this limitation, this study proposes a novel model termed Intensive Neural Learning with Classification Logic (INLCL) for classifying ultrasound images into benign and malignant categories. The proposed INLCL model is designed with a reduced number of trainable parameters to optimize performance while minimizing computational overhead. Key parameters-such as the number of filters, filter size, batch normalization, learning rate, number of epochs, and batch size-are carefully tuned to enhance accuracy while maintaining efficiency. To evaluate the effectiveness of the proposed model, its performance is compared against a standard classification technique, the Support Vector Machine (SVM), through cross-validation. The study utilizes a comprehensive dataset comprising 698 benign and 437 malignant cases, with ultrasound images annotated by experienced medical professionals. Extensive experimental results demonstrate that the INLCL model outperforms traditional approaches, including SVM, in terms of accuracy and other performance metrics, indicating its potential as an efficient and reliable tool for breast cancer detection using ultrasound imaging.