سامانه همایش‌ها TSTA | ثبت‌نام و ارسال مقاله کنفرانس‌ها و کنگره‌های ملی

صفحه اصلی / سی و دومین کنفرانس ملی و دهمین کنفرانس بین المللی مهندسی زیست پزشکی ایران

Investigating the Self-optimizing nnU-NetV2 for Kidney Tumor Segmentation: Application to the KiTS23 Dataset

نویسندگان :

Sanam Doostinia¹ Masoud Noroozi² Mohammad Saber Azimi³ Jafar Majidpour⁴ Hossein Arabi⁵

1- دانشگاه صنعتی خواجه نصیرالدین طوسی 2- دانشگاه اصفهان 3- دانشگاه شهید بهشتی 4- University of Raparin Rania, Iraq 5- Geneva University Hospital

کلمات کلیدی :

Deep Learning،Medical Image Segmentation،nnU-NetV2،Kidney Tumor Segmentation

چکیده :

Kidney cancer ranks among the top 10 most prevalent cancers, with renal cell carcinoma (RCC) being the dominant form, accounting for approximately 90% of all kidney cancer cases. As computer technology advances unprecedentedly, its integration into the medical field, particularly in computer-aided diagnostics and treatment, has grown significantly. In this work, we evaluate the nnU-NetV2 segmentation model on the kidney tumor segmentation dataset (KiTS2023). The 3D nnU-NetV2 model was trained for 300 Epochs with single-fold validation, using 320 CT scans from the retrospective KiTS23 dataset, with 80 cases for validation and 89 cases for testing. The evaluation metrics Dice Similarity Coefficient (DSC), IoU (Intersection over Union), sensitivity, and specificity were applied to assess performance in both region-based and foreground segmentation. Test-set DSC values were 0.8334 (Kidney+Tumor+Cyst), 0.6678 (Tumor+Cyst), and 0.6009 (Tumor); IoU scores were 0.7705, 0.5621, and 0.5078, respectively. Sensitivity values were 0.7915, 0.6743, and 0.6459, respectively, and specificity remained consistently high at 0.99 across all regions. For foreground segmentation on the test set, DSC was 0.7007, and IoU was 0.6135. Despite using a relatively low number of epochs and single-fold validation, comparison with the benchmark results demonstrates that the nnU-Net model remains a robust tool for automatic kidney tumor segmentation.