以一個先進的深度學習系統去分層高危險甲狀腺結節

Abstract

目的：目前甲狀腺結節使用超音波掃描分類系統去做分析非常耗時耗力且主觀。人工智慧系統（Artificial intelligence; AI）已被可以使用來增加預測甲狀腺結節惡性率的精準度。這研究主要目的是想要用目前最先進的Swin 變形分類器去對甲狀腺結節做分類。 方法：前瞻性地收集了臺中榮民總醫院2019年1月至2021年6月接受甲狀腺結節細針穿刺的超音波影像。總共有139位病患有甲狀腺惡性結節，有 235位病患是良性的甲狀腺結節當對照組。這些收集的超音波影像會用Swin-T 和ResNeSt50模組去對甲狀腺結節做分析。 結果：相較於良性結節的甲狀腺病患，有甲狀腺惡性結節的病患傾向於較年輕且男性居多。Swin-T的平均敏感度跟特異度分別為82.46%和84.29%。ResNeSt50的平均敏感度跟特異度分別為72.51%和77.14%。接受者操作特性曲線（ROC）分析指出曲線下方的面積（AUC），Swin-T為0.91，高於ResNeSt50曲線下方的面積0.82。麥內瑪檢定評估Swin-T和ResNeSt50的表現發現Swin-T有意義的比ResNeSt50的表現好。 結論：Swin-T 分類器對於醫師和有任何大小甲狀腺結節的病患，協助醫病共享決策是一個有用的工具，特別是在超音波影像特徵有高度惡性風險的甲狀腺結節。Swin-T 分類器對於分類甲狀腺結節，相較於目前先進的影像相似演算法也有著一致性的較佳的表現。

Purpose: The current ultrasound scan classification system for thyroid nodules is time-consuming, labor-intensive and subjective. Artificial intelligence (AI) has been shown to increase the accuracy of predicting the malignancy rate of thyroid nodules. This study aims to demonstrate the state-of-the-art Swin Transformer classifier to classify thyroid nodules. Methods: Ultrasound images were collected prospectively from patients who received fine-needle aspiration biopsy for thyroid nodules at Taichung Veterans Hospital, Taichung, Taiwan, from January 2019 to June 2021. A total of 139 patients with malignant thyroid nodules were enrolled while 235 patients with benign nodules served as controls. Images were fed to Swin-T and ResNeSt50 model to classify the thyroid nodules. Results: Patients with malignant nodules tend to be younger and male gender compared to those with benign nodules. The average sensitivity and specificity of Swin-T were 82.46% and 84.29%, respectively. The average sensitivity and specificity of ResNeSt50 were 72.51% and 77.14%, respectively. Receiver operating characteristics (ROC) analysis revealed that the area under the curve (AUC) of Swin-T was higher (AUC=0.91) than that of ResNeSt50 (AUC=0.82). The McNemar test evaluating performance between Swin-T and ResNeSt50 showed that Swin-T had significantly better performance than ResNeSt50. Conclusion: Swin-T classifier can be a useful tool in helping shared decision making between physicians and patients with thyroid nodules of any size, particularly in those with high-risk characteristics of sonographic patterns. The Swin-T classifier had consistently better performance classifying thyroid nodules when compared with current state-of-the-art image similarity algorithm.