کنفرانس مهندسی برق ایران

صفحه اصلی / سی و چهارمین کنفرانس بین المللی مهندسی برق

CEU-Net: A Connected EfficientUNet with ASPP for Mammographic Lesion Segmentation

نویسندگان :

Fatemeh Jafari¹ Saeed Meshgini² Reza Afrouzian³

1- دانشکده مهندسی برق و کامپیوتر دانشگاه تبریز 2- دانشکده مهندسی برق و کامپیوتر دانشگاه تبریز 3- دانشکده مهندسی برق و کامپیوتر دانشگاه تبریز

کلمات کلیدی :

Breast Cancer،segmentation،mammography،Unet

چکیده :

Breast cancer stands out as the most frequent and dangerous type of cancer in the female population due to the widespread development of cancer cells in the breast tissues, thereby causing detrimental health issues. Early detection using imaging technologies and medical examinations significantly improves treatment rates and reduces deaths. Imaging technologies, especially mammography, have long remained the fundamental diagnostic tool; nevertheless, accurate segmentation of cancer lesions in such images poses challenges due to the need for vast amounts of annotated data, the complex structure of breast tissue, and limitations of CNN models. To overcome such difficulties in mammographic image segmentation, an innovative segmentation model is proposed by combining two Efficient-UNets. In the model, the first network learns preliminary features, and the output of the smallest size after several convolutional operations is passed through the Atrous Spatial Pyramid Pooling (ASPP) module to generate features at multiple scales. These characteristics are further delivered to the Decoder part as inputs in the second network to ensure efficient rebuilding or segmentation of cancer-suspecting areas. Using such models effectively helps define boundaries around cancer lesions, especially in small regions, resulting in Dice and ACC scores of 91.00% and 92.24%, respectively.