BUS‐CAD: A computer‐aided diagnosis system for breast tumor classification in ultrasound images using grid‐search‐optimized machine learning algorithms with extended and Boruta‐selected features

Abstract

Breast cancer has become the most prominent type of cancer in the world. Early detection of breast cancer plays an important role in optimal treatment planning to decrease mortality. Breast ultrasound is widely used in diagnosing breast masses. Applications of machine learning in ultrasound imaging‐based classification have shown promising potential for early and accurate detection of breast cancer. In this study, a new computer‐aided diagnosis system based on machine learning techniques for breast cancer classification is proposed. Feature space is extended by using hybrid feature representations that combine both global and local texture statistics. A two‐step feature selection process is implemented using Boruta all‐relevant feature selection algorithm and iterative correlation analysis. A grid‐search strategy is followed along with 20 times repeated 10‐fold randomly stratified cross‐validation to optimize machine learning algorithms. Fourteen classification models based on random forest (RF) and support vector machine trained using all combinations of global features and the features driven from gray‐level co‐occurrence matrix and local binary patterns are tested. The experiments showed that the RF classifier on the hybrid feature vector that combines all global and local features achieved the best classification performance with average accuracy and area under the curve of 97.81% and 99.80%, respectively. The results suggest that the proposed system efficiently improves the classification performance of breast lesions on ultrasound images and can assist clinical decision‐making.

Graph neural network-based breast cancer diagnosis using ultrasound images with optimized graph construction integrating the medically significant features

“…BUSI A multiscale CNN classification model comprising 21 layers for classification, an autoencoder-based u-shaped DDA-Net segmentation model 97.89% 2. UDIAT Özcan ( 2023 ) BUSI BUS − CAD. Hybrid feature representation with global and local texture statistics, feature selection.…”

“…In addition, the time complexity for these models is much higher than our proposed model. (Özcan 2023 ) introduced a BUS−CAD system that includes global and textural-based feature statistics, feature selection and machine learning classification, achieved an accuracy of 97.81%. When compared with the test accuracies of the prior studies, it is evident that our proposed model has obtained the highest accuracy of 99.48%.…”

A novel fusion framework of deep bottleneck residual convolutional neural network for breast cancer classification from mammogram images

“…However, they employed the transfer learning idea and concentrated on the pre-trained models (i.e., VGG16 (24), Alexnet (25), ResNet (24), MobileNet (26), and EfficienctNet ( 27)). For training purposes, those models need a large and well-balanced dataset of images; however, the publicly accessible breast cancer dataset is insufficient (28). In addition, the extraction of irrelevant feature extraction decreased the classification accuracy and increased the computation time (second).…”