Paper Title
Relational-Based Knowledge Distillation With Multi-Layer Cross-Attentionin Ultrasound Elastography Cervical Lymph Node Image Classification

Abstract
Lymph nodes play a crucial role in the immune system, defending against foreign viruses and bacteria. Diagnosing neck lymph node tumors traditionally requires time-consuming processes, relying heavily on doctors' assessments of patients' lifestyle habits and physical examinations. Elastography ultrasound imaging has emerged as an alternative, enabling initial imaging examinations to guide the need for further diagnosis. However, the indistinct features in elastography ultrasound images complicate doctors' evaluations. To address this, we employ deep learning to distinguish between benign and malignant tumors, enhancing diagnostic accuracy. Utilizing a self-supervised learning approach for pretraining and improving model accuracy through our proposed knowledge distillation mechanism, our method demonstrates significant efficacy. On the neck elastography ultrasound imaging dataset from Far Eastern Memorial Hospital, our model achieves a sensitivity of 73.17%, specificity of 86.06%, and accuracy of 82.80%, accurately classifying benign and malignant tumors in ultrasound images. Keywords - Cervical Lymph Node, Ultrasound Elastography, Nowledge Distillation, Image Classification.