• NAVROOP GILL
• KAMAL KUMAR

Breast cancer is the second most common type of cancer after lung cancer. Early detection using screening mammography is instrumental in improving the breast cancer prognosis. Lesions in the breast are important signs of cancer and it is a challenging task to segment them with precision even with advanced deep learning methods. In this manuscript, we modify the cutting-edge CNN architecture, U-Net developed for biomedical image segmentation. We replace the encoder segment of the U-Net architecture with scalable and efficient Efficientnetb3 network. The model learns the local and global context in the input mammogram images and predicts a precise pixel-wise segmentation map for the lesion. These detected maps can assist the radiologists in reviewing the lesion location and shape in order to differentiate between benign and malignant lesions. We train and evaluate our model on the full mammographic (MG) scan images from the publicly available curated DDSM dataset (CBIS-DDSM). Our model yielded mean Intersection over Union (IoU) of 0.956 and Dice similarity coefficient of 0.968 between the radiologist marked ground truth masks (GTMs) and the segmented lesion maps generated by our model. This study aims to show that utilizing pre-trained encoder backbones in U-Net can result in faster convergence and high precision in lesion detection and segmentation tasks.

EfficientNet, FCN, image segmentation; transfer learning, Deep Learning (DL), convolutional neural networks (CNNs), mammogram, computer vision, U-Net, breast cancer detection

"Deep Learning Model for Breast Cancer Segmentation", International Journal of Emerging Technologies and Innovative Research (www.jetir.org), ISSN:2349-5162, Vol.8, Issue 9, page no.e117-e124, September 2021, Available :http://www.jetir.org/papers/JETIR2109518.pdf

"Deep Learning Model for Breast Cancer Segmentation", International Journal of Emerging Technologies and Innovative Research (www.jetir.org | UGC and issn Approved), ISSN:2349-5162, Vol.8, Issue 9, page no. ppe117-e124, September 2021, Available at : http://www.jetir.org/papers/JETIR2109518.pdf