Optimized Machine Learning for Automated Breast Ultrasound Segmentation: Improving Efficiency and Accuracy in Breast Cancer Diagnosis
By: Aleyna Isik, Young So
Departments: Biology, Computer Science
Faculty Advisor: Dr. Ilmi Yoon
Breast cancer is a widespread disease that requires timely detection and precise diagnosis for effective treatment and better outcomes. Various imaging techniques, such as mammography, MRI, and ultrasound, are commonly used to detect and diagnose breast cancer. Although ultrasound has advantages such as low cost, non-invasiveness, and lack of ionizing radiation exposure, interpreting ultrasound images can be challenging and time-consuming due to the complex nature of breast tissue. Automated segmentation of breast ultrasound images can aid in early detection, reduce inter-observer variability, and increase diagnostic efficiency. This study aimed to develop a model that accurately predicts the correct mask for each image, distinguishing between benign and malignant breast ultrasound images. The dataset was divided into a training set, validation set, and test set to train the model, optimize its hyperparameters, and evaluate its performance, respectively. Training time is critical in machine learning in medical imaging, particularly for large datasets and complex models. This study utilized an optimized environment tailored to the specific machine learning task, using ARM packages and utilities for Tensorflow, such as tensorflow-deps, tensorflow-metal, and tensorflow-macos, to accelerate the training process. The model achieved a dice score of over 0.75 on the test set, demonstrating the potential of machine learning for breast ultrasound segmentation. The results emphasize the importance of automating breast cancer diagnosis and the potential of machine learning to improve the accuracy and efficiency of diagnosis. Automated breast ultrasound segmentation can reduce subjectivity and variability in the diagnosis of breast cancer, identifying small or subtle lesions that may otherwise go undetected. The use of optimized environments and methodologies can further improve the efficiency and productivity of researchers and practitioners in this field.