Benign and malignant detection and classification for small size image of breast tumor recognition system using u-net model

• Main Authors: Suryanti, Awang, Kumar, Saumya, Nur Syafiqah, Mohd Nafis, Raihanah, Haroon
• Format: Conference or Workshop Item
• Language: English; English
• Published: IEEE 2024
• Subjects: QA75 Electronic computers. Computer science
• Online Access: http://umpir.ump.edu.my/id/eprint/43028/2/Benign%20and%20Malignant%20Detection%20and%20Classification%20for%20Small%20Size%20Image%20of%20Breast%20Tumor%20Recognition%20System%20using%20U-Net%20Model%20-%20intro.pdf; http://umpir.ump.edu.my/id/eprint/43028/1/Benign%20and%20Malignant%20Detection%20and%20Classification%20for%20Small%20Size%20Image%20of%20Breast%20Tumor%20Recognition%20System%20using%20U-Net%20Model.pdf; http://umpir.ump.edu.my/id/eprint/43028/; https://doi.org/10.1109/AiDAS63860.2024.10730254
• Institution: Universiti Malaysia Pahang Al-Sultan Abdullah

Breast tumor recognition is a critical task in the field of medical imaging systems, aiming to differentiate between benign and malignant tumors. To differentiate the tumors, an efficient technique is crucial to detect and classify it to avoid misdetection and misclassification, at the same time can accelerate the process. Thus, this paper proposed a deep learning technique which is a modified architecture of U-net model that based on Convolutional Neural Network (CNN) to detect and classify the tumors. The aim is to have a less complex U-Net model that is effective for a small size of images. During the technique deployment, data augmentation, transfer learning, and ensemble approach are employed. The proposed technique is tested using Breast Ultrasound Images dataset (BUSI) that is available in Kaggle. The results obtained are promising with accuracy of 0.8, precision of 0.88, recall of 0.7, and F1-score of 0.8. It indicates that this technique can contribute to the advancement of breast tumor detection and classification by providing valuable insights for clinicians in making accurate and timely diagnoses. Thus, the proposed technique has the potential to improve the efficiency and effectiveness of breast tumor recognition, aiding in the early detection and treatment of breast cancer.