Label-Free Rapid Intelligent Diagnosis of Thyroid Cancer

Abstract

The annual thyroid cancer incidence has been increasing. Thyroid cancer is categorized as a malignant neoplasm within the endocrine system. Fine-needle aspiration cytology remains the benchmark for thyroid cancer detection; however, the accuracy of the procedure depends on the practitioner’s expertise. Numerous challenges are associated with this process, such as obtaining inadequate cellular samples, mispuncturing the target lesion, and collecting nonrepresentative cell samples. These issues hinder proper cellular evaluation and increase the likelihood of misdiagnosis, ultimately impacting patient outcomes and the treatment trajectory. This research primarily aims to improve the diagnostic accuracy of thyroid cancer by introducing an innovative computer-aided diagnostic tool that leverages advanced deep learning techniques. Second-harmonic microscopy enables the fine extraction of morphological characteristics of collagen fibers within thyroid tissues, revealing significant differences in the distribution and organization of collagen fibers between normal and malignant tissues. In this study, we quantified the morphological alterations of collagen fibers by initially analyzing second-harmonic generation (SHG) images through a collagen scoring system based on feature extraction via least absolute shrinkage and selection operator regression. Model efficacy was assessed using receiver operating characteristic curves. Furthermore, we classified normal and malignant thyroid tissues in the validation cohort through three distinct deep-learning architectures (Mobile Neural Networks Version 3 (MobileNetV3), Visual Geometry Group 16(VGG16), and Pyramid Vision Transformer v2 (PVTv2) in combination with SHG image data. Overall, MobileNetV3 achieved the best classification performance (87.4%). This study provides preliminary evidence for the effectiveness of deep-learning algorithms in differentiating between malignant and normal thyroid tissues. This significant advancement offers valuable technological support for detecting thyroid cancer in clinical environments and is expected to enhance both the accuracy and efficiency of diagnostic practices.