Dataset Information

Diagnosing thyroid nodules with atypia of undetermined significance/follicular lesion of undetermined significance cytology with the deep convolutional neural network.

ABSTRACT: To compare the diagnostic performances of physicians and a deep convolutional neural network (CNN) predicting malignancy with ultrasonography images of thyroid nodules with atypia of undetermined significance (AUS)/follicular lesion of undetermined significance (FLUS) results on fine-needle aspiration (FNA). This study included 202 patients with 202 nodules ≥ 1 cm AUS/FLUS on FNA, and underwent surgery in one of 3 different institutions. Diagnostic performances were compared between 8 physicians (4 radiologists, 4 endocrinologists) with varying experience levels and CNN, and AUS/FLUS subgroups were analyzed. Interobserver variability was assessed among the 8 physicians. Of the 202 nodules, 158 were AUS, and 44 were FLUS; 86 were benign, and 116 were malignant. The area under the curves (AUCs) of the 8 physicians and CNN were 0.680-0.722 and 0.666, without significant differences (P > 0.05). In the subgroup analysis, the AUCs for the 8 physicians and CNN were 0.657-0.768 and 0.652 for AUS, 0.469-0.674 and 0.622 for FLUS. Interobserver agreements were moderate (k = 0.543), substantial (k = 0.652), and moderate (k = 0.455) among the 8 physicians, 4 radiologists, and 4 endocrinologists. For thyroid nodules with AUS/FLUS cytology, the diagnostic performance of CNN to differentiate malignancy with US images was comparable to that of physicians with variable experience levels.

SUBMITTER: Youn I

PROVIDER: S-EPMC8501016 | biostudies-literature | 2021 Oct

REPOSITORIES: biostudies-literature

ACCESS DATA

Publications

Diagnosing thyroid nodules with atypia of undetermined significance/follicular lesion of undetermined significance cytology with the deep convolutional neural network.

Youn Inyoung I Lee Eunjung E Yoon Jung Hyun JH Lee Hye Sun HS Kwon Mi-Ri MR Moon Juhee J Kang Sunyoung S Kwon Seul Ki SK Jung Kyong Yeun KY Park Young Joo YJ Park Do Joon DJ Cho Sun Wook SW Kwak Jin Young JY

Scientific reports 20211008 1

To compare the diagnostic performances of physicians and a deep convolutional neural network (CNN) predicting malignancy with ultrasonography images of thyroid nodules with atypia of undetermined significance (AUS)/follicular lesion of undetermined significance (FLUS) results on fine-needle aspiration (FNA). This study included 202 patients with 202 nodules ≥ 1 cm AUS/FLUS on FNA, and underwent surgery in one of 3 different institutions. Diagnostic performances were compared between 8 physicians ...[more]

PMID: 34625636

Similar Datasets

Project description:BackgroundMicroRNA (miR) expression signatures are proposed to be able to differentiate thyroid cancer from benign thyroid lesions. We selected eight miRs (miR-146b, -221, -187, -197, -346, -30d, -138, and -302c) to examine the potential use of miRs to supplement diagnostic cytology in cases designated as "atypia of undetermined significance."MethodsmiR expression was measured in thyroid fine needle aspiration (FNA) specimens by quantitative polymerase chain reaction. Gene expression analyses and linear discriminant analysis (LDA) were performed in a training sample set (n=60) to obtain a classification rule to predict FNA cases as benign or malignant. The predictions were cross-validated by comparing with the corresponding histological diagnoses. A validation sample set (n=68) was further tested with the established four-miR LDA classification rule.ResultsA set of four miRs (miR-146b, -221, -187, and -30d) was identified that could differentiate malignant from benign lesions. A four-miR LDA classification rule was obtained and used to predict FNA cases as benign or malignant. For the training sample set, we obtained a diagnostic accuracy of 93.3%, sensitivity of 93.2%, specificity of 93.8%, positive predictive value (PPV) of 0.98, and negative predictive value (NPV) of 0.83. For the validation sample set, we obtained a diagnostic accuracy of 85.3%, sensitivity of 88.9%, specificity of 78.3%, PPV of 0.89, and NPV of 0.78. For the 30 atypia cases in the validation sample set, we obtained a diagnostic accuracy of 73.3%, sensitivity of 63.6%, specificity of 78.9%, PPV of 0.64, and NPV of 0.79. Based on the miR predictions, we classified the atypia cases predicted as "malignant" into "high risk" and those predicted as "benign" into "low risk" categories. While thyroid carcinomas, particularly papillary thyroid carcinomas (PTCs), were relatively enriched in the high-risk category, this particular miR panel is subject to inaccurate results in follicular neoplasias in atypia cases.ConclusionsWe demonstrate that miR amplification from FNA samples is feasible and that the particular four miR profile in this study can identify PTCs. However, further refinement is required for application to FNA cytology of "atypia of undetermined significance" cases due to low accuracy in classifying follicular neoplasias.

Project description:BackgroundFine-needle aspiration (FNA) cytology is a common approach to evaluating thyroid nodules, although 20% to 30% of FNAs have indeterminate cytology, which hampers the appropriate management of these patients. Follicular (or oncocytic) neoplasm/suspicious for a follicular (or oncocytic) neoplasm (FN/SFN) is a common indeterminate diagnosis with a cancer risk of approximately 15% to 30%. In this study, the authors tested whether the most complete next-generation sequencing (NGS) panel of genetic markers could significantly improve cancer diagnosis in these nodules.MethodsThe evaluation of 143 consecutive FNA samples with a cytologic diagnosis of FN/SFN from patients with known surgical outcomes included 91 retrospective samples and 52 prospective samples. Analyses were performed on a proprietary sequencer using the targeted ThyroSeq v2 NGS panel, which simultaneously tests for point mutations in 13 genes and for 42 types of gene fusions that occur in thyroid cancer. The expression of 8 genes was used to assess the cellular composition of FNA samples.ResultsIn the entire cohort, histologic analysis revealed 104 benign nodules and 39 malignant nodules. The most common point mutations involved the neuroblastoma RAS viral oncogene homolog (NRAS), followed by the Kirsten rat sarcoma viral oncogene homolog (KRAS), the telomerase reverse transcriptase (TERT) gene, and the thyroid-stimulating hormone receptor (TSHR) gene. The identified fusions involved the thyroid adenoma associated (THADA) gene; the peroxisome proliferator-activated receptor γ (PPARG) gene; and the neurotrophic tyrosine kinase, receptor, type 3 (NTRK3) gene. Performance characteristics were similar in the retrospective and prospective groups. Among all FN/SFN nodules, preoperative ThyroSeq v2 performed with 90% sensitivity (95% confidence interval [CI], 80%-99%), 93% specificity (95% CI, 88%-98%), a positive predictive value of 83% (95% CI, 72%-95%), a negative predictive value of 96% (95% CI, 92%-100%), and 92% accuracy (95% CI, 88%-97%).ConclusionsThe current results indicate that comprehensive genotyping of thyroid nodules using a broad NGS panel provides a highly accurate diagnosis for nodules with FN/SFN cytology and should facilitate the optimal management of these patients.

Dataset Information

Diagnosing thyroid nodules with atypia of undetermined significance/follicular lesion of undetermined significance cytology with the deep convolutional neural network.

Publications

Diagnosing thyroid nodules with atypia of undetermined significance/follicular lesion of undetermined significance cytology with the deep convolutional neural network.

Similar Datasets

OmicsDI is part of the ELIXIR infrastructure

Tweets