Automatic Segmentation of Cancerous Tissues in Breast Ultrasound Images

Abstract

Segmentation of cancerous tissues in breast ultrasound images is a challenging task that can be
achieved through various image processing and machine learning techniques. This paper presents an
early automatic diagnostic method for breast cancer in ultrasound images and classifying the cancer as benign
or malignant. Automatic images processing-based techniques were applied to process segment and
classify cancerous tissues. The proposed method for early detection of breast tumor in ultrasound images
typically involves three key steps: preprocessing, segmentation, and feature extraction and
classification. The accuracy of proposed method was evaluated by comparing the segmented images with
the gold standard or ground truth (GT), which was manually formed by an experienced radiologist. Areabased evaluation of segmentation was performed in
which root mean squared error was 0.0435 and relative absolute error was 0.2085% for the total instances. The
true positive result was 96% while false positive result was 3.11 %. The early detection of breast cancer in
ultrasound images using the proposed image processing-based technique accurately detect breast
cancer and classify cancer as benign or malignant. It also eliminates the drawbacks in the existing methods
by introducing new features like shadows in the images using masking and adjusting contrast
functions.