超声医生如何说？| AI安克侦可辅助超声医师提高对甲状腺结节的诊断能力

随着人们健康意识的提升、超声诊断技术的发展，甲状腺占位的发病率逐年上升，甲状腺癌已成为美国女性第5大常见疾病，也是很多国家最常见癌症之一。超声是甲状腺占位性病变的主要影像学检查，而常规超声诊断主要依靠医师视觉与经验判断，一方面因视觉局限而获取的信息不准确导致诊断准确性具有瓶颈，另一方面，对超声医师的经验具有一定要求，因此，一直以来成为常规超声诊断甲状腺占位病变的一个瓶颈和挑战。

由中国人民解放军总医院的Dr. YuanYuan Lu 及Prof. Junlai Li在国际权威期刊《Journal of Ultrasound in Medicine》发表最新临床研究,题为《Value of Computer Software for Assisting Sonographers in the Diagnosis of Thyroid Imaging Reporting and Data System Grade 3 and 4 Thyroid Space-Occupying Lesions》。译：计算机软件辅助超声医师诊断甲状腺TI-RADS 3~4级占位病变的价值。

本研究使用专业的甲状腺超声图像处理系统（安克侦 AmCAD-UT Detection，由中国台湾的安克生医股份有限公司提供），能够自动量化甲状腺结节区域的超声特征，并用彩色突出增强显示关键特征，帮助医师更清晰地观察超声图像；并且依据多个国际甲状腺超声诊断指南自动计算出分级或评分，供医师做诊断参考。通过此项临床研究证实：本研究使用的安克侦CAD软件可有效辅助超声医师提高对甲状腺TI-RADS 3~4级疑难占位病变的诊断能力；CAD的诊断建议与穿刺活检的结果相对一致，有助于降低不必要的穿刺活检比例。

目的：分析甲状腺超声的计算机辅助诊断软件（安克侦）在辅助医师诊断甲状腺TI-RADS 3~4级占位病变的能力，并为超声医生使用此系统的诊断建议进行临床决策支持提供证据。

方法：在第1组中，对234例手术病理证实的甲状腺病变进行了回顾性研究。对比同一医师在使用该软件前、结合软件后及软件本身独立诊断三者对甲状腺占位病例诊断的灵敏度、特异度及准确率差异。在第2组中，对220名患有甲状腺占位性病变的病例进行了一项前瞻性研究，该病例由我院医生推荐，并实施穿刺活检进行确诊。将超声图像导入安克侦软件（AmCAD-UT）中，并获得软件给出的诊断建议：建议穿刺活检或者定期随访。根据穿刺活检的病理结果，计算出安克侦软件（AmCAD-UT）诊断建议的一致性和符合率。

结果：在第1组中，CAD软件独立和超声医生结合CAD的诊断都比超声医生独立诊断获得更高的敏感性和诊断准确性（P <0.05）。在第2组中，软件显示整体组内相关性（κ= 0.786），并且与穿刺活检的诊断结果符合率为93.6％。

结论：安克侦软件（AmCAD-UT）可辅助超声医师提高对甲状腺TI-RADS 3~4级占位病变的诊断能力。AmCAD的诊断建议与穿刺活检的结果相对一致，有助于降低不必要的穿刺活检比例。

Objectives—To analyze the ability of thyroid ultrasound computer-aided diagnosis (CAD) detection software (AmCAD-UT; AmCAD BioMed Corporation, Taipei, Taiwan) to assist sonographers in diagnosing Thyroid Imaging Reporting and Data System grade 3 and 4 space-occupying lesions and to provide evidence for ultrasound doctors (UDs) to use the diagnostic recommendations of the AmCAD system to inform clinical decisions.

Methods—In group 1, a retrospective study was performed on 234 cases of thyroid lesions confirmed by surgical pathology. The sensitivities, specificities, and accuracies of the diagnoses determined by the same Sonographers   independent of the software (UD) and after consulting the CAD software (UD+ CAD) and by the software alone (CAD) were compared. In group 2, a prospective study was performed on 220 individuals with thyroid space-occupying lesions recommended by physicians from our hospital to undergo needle biopsy to confirm the diagnosis. Ultrasound images were imported into AmCAD, and recommendations for needle biopsy or periodic follow-up were obtained. According to the pathologic results of needle biopsy, consistency and coincidence rates of diagnostic recommendations for AmCAD were obtained.

Results—In group1,CAD and UD+CAD diagnoses achieved significantly higher sensitivities and accurancies of diagnosis than did independent diagnosis by the UD (P < .05). In group 2, the software showed an overall intraclass correlation (κ = 0.786) and a diagnosis coincidence rate of 93.6% with needle biopsy results.

Conclusions—AmCAD-UT Detection improved the ability of UDs to diagnose Thyroid Imaging Reporting and Data System grade 3 and 4 space-occupying lesions. Diagnostic recommendations of AmCAD are relatively consistent with needle biopsy results and can reduce the rate of unnecessary diagnostic needle biopsies.