Scuola di Specializzazione in Radiodiagnostica, Università degli Studi di Milano, via Festa del Perdono 7, 20122, Milan, Italy.
Scuola di Specializzazione in Endocrinologia e Malattie del Metabolismo, Università degli Studi di Milano, Milan, Italy.
Radiol Med. 2019 Feb;124(2):118-125. doi: 10.1007/s11547-018-0942-z. Epub 2018 Sep 22.
To compare the diagnostic performance of a commercially available computer-aided diagnosis (CAD) system for thyroid ultrasound (US) with that of a non-computer-aided radiologist in the characterization of low-to-high suspicion thyroid nodules.
This retrospective study included a consecutive series of adult patients referred for US-guided fine-needle aspiration biopsy (FNAB) of a thyroid nodule. All patients were eligible for thyroid nodule FNAB according to the current international guidelines. An interventional radiologist experienced in thyroid imaging acquired the US images subsequently used for post-processing, performed FNAB and provided the US features of each nodule. A radiology resident and an endocrinology resident in consensus performed post-processing using the CAD system to assess the same nodule characteristics. The diagnostic performance and agreement of US features between the CAD system and the radiologist were compared.
Sixty-two patients (50 F; age 60 ± 12 years) were enrolled: 77.4% (48/62) of thyroid nodules were benign, 22.6% (14/62) were undetermined to malignant and required follow-up or surgery. Interobserver agreement between the CAD system and the radiologist was substantial for orientation (K = 0.69), fair for composition (K = 0.36), echogenicity (K = 0.36), K-TIRADS (K = 0.29), and slight for margins (K = 0.03). The radiologist demonstrated a significantly higher sensitivity than the CAD system (78.6% vs. 21.4%; P = 0.008), while there was no statistical difference in specificity (66.7% vs. 81.3%; P = 0.065).
This CAD system is less sensitive than an experienced radiologist and showed slight-to-substantial agreement with the radiologist for the characterization of thyroid nodules. Although it is an innovative tool with good potential, additional efforts are needed to improve its diagnostic performance.
比较一种商业化的计算机辅助诊断 (CAD) 系统与非计算机辅助的放射科医师在诊断低至高可疑甲状腺结节中的表现。
这是一项回顾性研究，纳入了连续系列因甲状腺超声（US）引导下细针抽吸活检（FNAB）而转诊的成年患者。所有患者均符合当前国际指南对甲状腺结节 FNAB 的适应证。一位具有甲状腺影像学经验的介入放射科医师获取了 US 图像，随后进行了后处理，并进行了 FNAB 检查，提供了每个结节的 US 特征。一名放射科住院医师和一名内分泌科住院医师使用 CAD 系统进行共识后处理，以评估同一结节特征。比较 CAD 系统和放射科医师的 US 特征的诊断性能和一致性。
共纳入 62 名患者（50 名女性；年龄 60±12 岁）：77.4%（48/62）的甲状腺结节为良性，22.6%（14/62）为不确定为恶性，需要随访或手术。CAD 系统与放射科医师在方位（K=0.69）、成分（K=0.36）、回声（K=0.36）、K-TIRADS（K=0.29）和边缘（K=0.03）方面具有显著的一致性。放射科医师的敏感性显著高于 CAD 系统（78.6%比 21.4%；P=0.008），而特异性无统计学差异（66.7%比 81.3%；P=0.065）。
与经验丰富的放射科医师相比，该 CAD 系统的敏感性较低，且在甲状腺结节的特征描述方面与放射科医师具有轻度至中度一致性。尽管它是一种具有良好潜力的创新工具，但仍需进一步努力提高其诊断性能。
