Urfali, AtaberkUymaz, Sait Ali2025-10-102025-10-1020251863-17031863-1711https://doi.org/10.1007/s11760-025-04629-zhttps://hdl.handle.net/20.500.13091/10866Head and neck cancers are of paramount importance due to their high prevalence and potential fatality, making early diagnosis and precise treatment planning crucial. Accurate identification and segmentation of OARs is essential in minimizing radiation exposure to healthy tissues during treatment. This study introduces NanoUnet, a novel deep learning network designed to overcome challenges in segmenting OARs from CT scans. NanoUnet employs depthwise separable convolutions, reducing computational complexity by 95.7% while maintaining high performance, achieving a Dice score of 0.981 for larger structures such as the brainstem and 0.803 for smaller, more complex structures. Additionally, SE blocks are integrated to enhance segmentation accuracy through focused feature recalibration, leading to improved performance across 24 different OAR classes, including critical structures like the brain, spinal cord, and optic nerves. Evaluated on the Segrap 2023 dataset, NanoUnet demonstrated superior segmentation accuracy, making it well-suited for clinical applications. These findings support NanoUnet's potential to enhance treatment planning accuracy and efficiency for head and neck cancer patients, contributing to the field's advancement towards personalized and precise radiotherapy.eninfo:eu-repo/semantics/closedAccessNasopharyngeal CarcinomaSEGRAP 2023Deep LearningU-NetOrgans at RiskGross Tumor VolumeArticle10.1007/s11760-025-04629-z2-s2.0-105015358212