کنفرانس مهندسی برق ایران

صفحه اصلی / سی و چهارمین کنفرانس بین المللی مهندسی برق

Combination of a Convolutional Neural Network and a Vision Transformer for Segmenting Internal Organs in CT Scans

نویسندگان :

Mohammad Amin Nooraei Yeganeh¹ Hamid Soltanian-Zadeh²

1- دانشگاه تهران، دانشکده مهندسی برق و کامپیوتر 2- دانشگاه تهران، دانشکده مهندسی برق و کامپیوتر

کلمات کلیدی :

Organ Segmentation،Computed Tomography (CT)،Convolutional Neural Networks (CNNs)،Vision Transformers،Hybrid Deep Learning Models

چکیده :

Accurate segmentation of abdominal organs in CT scans is essential for diagnosis, treatment planning, and modern computer-assisted clinical systems. In this study, we introduce a hybrid framework that combines the strengths of convolutional networks and vision transformers to better capture both local structural details and long-range spatial relationships. The proposed model incorporates EfficientNetV2 for multi-scale feature representation and a BiFormer Transformer module to enhance global context modeling, enabling more reliable delineation of organ boundaries in complex CT images. The method was evaluated on the Synapse multi-organ CT dataset, where it achieved an average Dice score of 81.08%, outperforming several recent state-of-the-art segmentation models. Despite its strong performance, the model remains computationally efficient, requiring only 20.82M parameters and 5.12 GFLOPs. These results highlight the effectiveness of integrating CNN-based local feature extraction with transformer-based global reasoning and demonstrate the potential of the proposed approach for clinical-grade abdominal organ segmentation.