Texture Analysis for Breast Ultrasound Using Conventional Method and Deep Learning
Ruey‐Feng Chang, Yao‐Sian Huang, Yan‐Wei Lee · 2024
Early detection and diagnosis are the best way to improve patient prognosis in breast cancer. Computer-aided diagnosis (CAD) systems have been a useful tool for reducing oversight error and raising cancer diagnosis accuracy. In this chapter, there are two CAD systems with conventional and novel schemes proposed to overcome two issues. One issue is the classification between triple-negative breast cancer (TNBC) and fibroadenoma because of the similarity of morphology features on breast ultrasound. Another issue is how to develop a breast cancer CAD system in which the used dataset is collected from several different US vendors due to the inconsistency of US machines. Firstly, a CAD system is developed with the morphology, the conventional texture, and the multi-resolution invariant texture features for fibroadenoma and TNBC classification. Secondly, an auto-learned CAD system is trained for overcoming the inconsistency problem of different machines. In this issue, not only clinically collected US images but also an open dataset (BUSI) are employed for validating system performance. In two issues, six performance indices—accuracy, sensitivity, specificity, positive predictive value (PPV), negative predictive value, and area under receiver operator characteristic curve (Az)—are computed to evaluate system performance.