Medical Image Segmentation using EfficientNet-based U-Net Architecture

Medical image segmentation plays a pivotal role in computer-aided diagnosis and treatmentplanning. This study focuses on improving the accuracy of medical image segmentation, a crucial stepin diagnosing and treating diseases. We explore the use of a powerful neural network, EfficientNet,combined with the U-Net architecture. We compare the performance of our model with a conventionalU-Net variant utilizing the ResNet۳۴ backbone, aiming to assess the efficacy of EfficientNet in thisdomain.Our experiments involve extensive training on a curated dataset, encompassing diverse anatomicalregions, including large bowel, small bowel, and stomach. We evaluate the segmentation performancethrough metrics such as Intersection over Union (IoU) and Dice Coefficient. Additionally, we discussthe importance of proper data preprocessing, including input normalization and augmentation, inachieving robust segmentation results.The presented EfficientNet-based U-Net architecture holds significant potential for real-worlddeployment in medical image analysis, offering improved segmentation accuracy and computationalefficiency. This work contributes to the ongoing exploration of deep learning architectures in medicalimaging, paving the way for advancements in clinical diagnosis and treatment planning.

کلیدواژه ها:

Medical imaging ، Segmentation ، U-Net Architecture ، EfficientNet