An artificial intelligence ultrasound system’s ability to distinguish benign from malignant follicular-patterned lesions

Abstract

ObjectivesTo evaluate the application value of a generally trained artificial intelligence (AI) automatic diagnosis system in the malignancy diagnosis of follicular-patterned thyroid lesions (FPTL), including follicular thyroid carcinoma (FTC), adenomatoid hyperplasia nodule (AHN) and follicular thyroid adenoma (FTA) and compare the diagnostic performance with radiologists of different experience levels.MethodsWe retrospectively reviewed 607 patients with 699 thyroid nodules that included 168 malignant nodules by using postoperative pathology as the gold standard, and compared the diagnostic performances of three radiologists (one junior, two senior) and that of AI automatic diagnosis system in malignancy diagnosis of FPTL in terms of sensitivity, specificity and accuracy, respectively. Pairwise t-test was used to evaluate the statistically significant difference.ResultsThe accuracy of the AI system in malignancy diagnosis was 0.71, which was higher than the best radiologist in this study by a margin of 0.09 with a p-value of 2.08×10-5. Two radiologists had higher sensitivity (0.84 and 0.78) than that of the AI system (0.69) at the cost of having much lower specificity (0.35, 0.57 versus 0.71). One senior radiologist showed balanced sensitivity and specificity (0.62 and 0.54) but both were lower than that of the AI system.ConclusionsThe generally trained AI automatic diagnosis system can potentially assist radiologists for distinguishing FTC from other FPTL cases that share poorly distinguishable ultrasonographical features.

Ultrasound characteristics of follicular and parafollicular thyroid neoplasms: diagnostic performance of artificial neural network

“…Untile today, most AI applications in thyroid disease have focused on estimation of the malignancy risk of nodules [ 20 ]. In a retrospective study of follicular thyroid nodules, Xu et al achieved an accuracy of 0.71 for discriminating between benign and malignant lesions by neural network analysis [ 21 ]. Hence, the potential benefit of this approach in helping radiologists separate FTCs from nodules poorly distinguishable by US was duly illustrated.…”

Differential regional importance mapping for thyroid nodule malignancy prediction with potential to improve needle aspiration biopsy sampling reliability

“…Good performance in distinguishing malignant from benign nodules also plays a vital role in identifying the "hot" regions in the malignancy heat map to be the key regions for the diagnosis of malignant nodules. Furthermore, other recent studies have shown that the AI system has balanced specificity and sensitivity with overall diagnostic accuracy matching high-performing senior radiologists (24), can outperform senior radiologists in diagnosing rare thyroid carcinomas (25), and can be potentially helpful for discrimination between malignant and benign follicular-patterned thyroid lesions (21). Interestingly, the shapes and subregions of cancer cell-rich pathological images showed good correspondence to the heat maps computed from the ultrasound images, suggesting a promising potential to use the heat map visualization to guide targeted FNAB for more reliable sampling compared with conventional ultrasound-guided sampling.…”

“…The CAM-based heat maps computed from AI-CADx systems with diagnostic performances comparable to or even better than those of senior radiologists (27-29) on ultrasound images with the capability of differentiating regional importance for malignancy diagnosis may provide additional guidance to localize diagnosisenabling nodular regions, especially large ones, for more accurate sampling, given that the number of needle passes has to be limited. In addition, FNAB-based cytopathological examination is acknowledged to have a limitation in diagnosing follicularpatterned thyroid lesions (FPTLs) (30-32), while the AI system was found to be 69% accurate in differentiating thyroid follicular carcinoma from benign FPTL cases (21), suggesting that the heat maps developed on top of the AI system may provide better guidance than plain ultrasound for FPTL sampling by FNAB to help with newly developed proteomics-based diagnosis (33). Of course, for smaller thyroid nodules, it may be difficult to precisely guide FNAB of the segmented nodular regions based on the malignancy heat map.…”

“…This study was based on the AI-SONIC ™ Thyroid nodule diagnosis system (Demetics Medical Technology, Hangzhou, China). It is built on the EfficientNet (20) architectural backbone and employs supervised sharpness-aware minimization for model parameter optimization to realize automatic nodule segmentation and classification (21). The classification module of the AI model relies on the precise localization of the nodules.…”

Multi-Omics and Management of Follicular Carcinoma of the Thyroid

“…An integration and concurrent analysis of findings from ultrasound and elastography, radioisotope scan and MRI is most likely going to increase the accuracy of predicting the likelihood of FTC. The extent of vascular or capsular invasion of the FTC into may also be determined without the need for an invasive biopsy or a costly and potentially risky diagnostic lobectomy [ 176 , 217 , 218 , 219 , 220 , 221 , 222 , 223 ] Radiomics enables segmentation of the tumor and facilitates a thorough analysis of the tumor and the TME [ 163 , 224 , 225 ].A thorough analysis of a thyroid nodule and the TME may enable a “virtual” biopsy of the lesion and accurate histological diagnosis and grading of the tumor, like what would be achieved following a tissue or liquid biopsy. Some of the imaging investigations, such as [ 18 F] FDG PET/CT and [ 18 F] FDG/MRI, can localize metastases from FTC when there is discordance between the serum Tg level and the results of a whole-body radioiodine scan (WBS) [ 72 , 131 , 142 , 226 ].…”