Segmentation and Evaluation of Breast Cancer in Ultrasound Images: A Study with UNet Models

Breast Cancer (BC) is a harsh disease in women and needs appropriate detection and handling. Untreated BC will lead to death and hence a number of diagnostic approaches are proposed to detect its harshness in early phase. Breast Ultrasound Imaging (BUI) is a noninvasive medical imaging modality to detect the BC and its harshness and it is a safe and low cost procedure. Due to its merit, it is widely adopted for clinical level screening of the BC. In this work, UNet-based segmentation is presented for analyzing the BC from the BUI images. The various phases of this work includes the following; (i) image and mask collection and resizing it to 224 × 224 × 1 pixels, (ii) training and verifying the performance of the segmentation scheme with image and mask, and (iii) testing the performance to confirm the clinical significance. This work considered the benchmark BUI data for the evaluation and the performance of the developed system is verified using the unprocessed and CLAHE processed images. The experimental outcome of this study confirms that the CLAHE enhanced BUI helps to provide an average accuracy of >95% with the UNetbased segmentation.