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

Enhancing Dental Disease Detection: Leveraging Swin Transformer and DenseNet with Attention-Guided Fusion in Dental Panoramic Imaging

نویسندگان :

Mahdieh Dehghani¹ Reza Aghaeizadeh Zoroofi²

1- School of Electrical and Computer Engineering College of Engineering,University of Tehran 2- School of Electrical and Computer Engineering College of Engineering,University of Tehran

کلمات کلیدی :

Dental Disease Detection،Panoramic Dental X-ray Images،DenseNet،Swin Transformer،Guided Fusion technique

چکیده :

The automatic identification of dental diseases and anomalies in panoramic radiographs poses a significant challenge, primarily due to the vast array of dental conditions that can manifest. The synergistic integration of multiple deep learning architectures has the potential to enhance overall performance, as each model compensates for the limitations present in the others. In the context of deep learning, the choice of backbone architecture is critical for effective feature extraction. Consequently, employing diverse feature extraction techniques can substantially augment the model's capability to accurately identify dental conditions. This paper introduces an approach for detecting diseases and abnormalities in panoramic dental images through the application of an Attention-Guided Fusion technique, which integrates features extracted from two distinct architectures: DenseNet and Swin Transformer. The DenseNet framework excels in capturing local features, while the Swin Transformer is proficient in obtaining global contextual information. By adaptively fusing these features, the proposed method effectively highlights critical regions within the images.The performance of the proposed model was rigorously evaluated using a publicly available dataset, demonstrating a significant improvement in accuracy compared to several established methodologies, including Faster R-CNN, Combined DINO and YOLOv8, DETR, YOLOrtho, HierarchicalDet, and DiffusionDet.