Method for MICCAI FLARE24 Challenge
Keywords: Cancer·Segmentation
Abstract: In recent years, the prevalence of cancer cases has been significant, with projections indicating a substantial increase by 2024. Effective radiation
therapy (RT) is crucial for treatment, yet it poses risks to surrounding organs. Accurate delineation of organs at risk (OARs) on CT images is essential to mitigate these risks. Medical imaging, particularly CT scans, plays a pivotal role in tumor diagnosis and treatment planning. Automated segmentation of
tumors in pan-cancer CT scans using advanced computational techniques, such
as deep learning, is pivotal for improving clinical decision-making. This paper explores a novel training strategy based on the ResUNet model to enhance
segmentation accuracy. The strategy involves phased training, first without skip
connections for primary segmentation and then with skip connections for detailed segmentation, aiming to improve both efficiency and precision. Experimental results demonstrate promising potential, highlighting the method's applicability and future directions in medical image segmentation.
Submission Number: 23
Loading