Computer-aided diagnosis (CAD) systems hold potential to improve the diagnostic accuracy of thyroid ultrasound (US). We aimed to develop a deep learning-based US CAD system (dCAD) for the diagnosis of thyroid nodules and compare its performance with those of a support vector machine (SVM)-based US CAD system (sCAD) and radiologists. dCAD was developed by using US images of 4919 thyroid nodules from three institutions. Its diagnostic performance was prospectively evaluated between June 2016 and February 2017 in 286 nodules, and was compared with those of sCAD and radiologists, using logistic regression with the generalized estimating equation. Subgroup analyses were performed according to experience level and separately for small thyroid nodules 1–2 cm. There was no difference in overall sensitivity, specificity, positive predictive value (PPV), negative predictive value and accuracy (all p > 0.05) between radiologists and dCAD. Radiologists and dCAD showed higher specificity, PPV, and accuracy than sCAD (all p < 0.001). In small nodules, experienced radiologists showed higher specificity, PPV and accuracy than sCAD (all p < 0.05). In conclusion, dCAD showed overall comparable diagnostic performance with radiologists and assessed thyroid nodules more effectively than sCAD, without loss of sensitivity.

Sentinel lymph node biopsy (SLNB) can be performed when node-positive disease is converted to node-negative status after neoadjuvant chemotherapy (NCT). Tattooing nodes might improve accuracy but supportive data are limited. This study aimed to investigate the feasibility of charcoal tattooing metastatic axillary lymph node (ALN) at presentation followed by SLNB after NCT in breast cancers. Twenty patientswith cytology-proven node metastases prospectively underwent charcoal tattooing at diagnosis. SLNB using dual tracers and axillary surgery after NCT were then performed. The detection rate of tattooed node and diagnostic performance of SLNB were analyzed. All patients underwent charcoal tattooingwithout significant morbidity. Sentinel and tattooed nodes could be detected during surgery after NCT. Nodal pathologic complete response was achieved in 10 patients. Overall sensitivity, false-negative rate (FNR), negative predictive value, and accuracy of hot/blue SLNB were 80.0%, 20.0%, 83.3%, and 90.0%, respectively. Retrieving more nodes and favorable nodal response were associated with improved performance. The best accuracy was observed when excised tattooed node was calculated together (FNR, 0.0%). Cold/non-blue tattooed nodes of five patients were removed during non-sentinel axillary surgery but clinicopathological parameters did not differ compared to patients with hot/blue tattooed node detected during SLNB, suggesting the importance of the tattooing procedure itself to improve performance. Charcoal tattooing of cytology-confirmed metastatic ALN at presentation is technically feasible and does not limit SLNB after NCT. The tattooing procedure without additional preoperative localization is advantageous for improving the diagnostic performance of SLNB in this setting.

Breast magnetic resonance (MR) imaging has high sensitivity in breast cancer diagnosis. The probability of malignancy for additional detected lesions is higher in patients with breast cancer than in the population without malignancy, which is why biopsy or further study should be considered for additional detected lesions. Because of the shortcomings of MR imaging-guided biopsy, second-look ultrasonography (US) may be the preferred next step. Detecting target lesions at second-look US and correlating lesions between the two modalities may be challenging. Using axial MR imaging to localize the lesion with respect to the nipple and the lesion-to-nipple distance can narrow the scan range at US. Evaluating the lesion's location relative to the mammary zones and surrounding tissues, as well as noting its depth, characteristics, and nearby landmarks, will aid in lesion correlation. Doppler imaging, tissue harmonic imaging, and other US techniques can be used to identify subtle lesions. Although malignant breast lesions may appear probably benign at second-look US, decision making for biopsy must be based primarily on MR imaging findings. In sonographically occult, MR imaging-detected lesions with suspicious MR imaging features, the probability of malignancy is much higher than 2%, and MR imaging-guided biopsy must be performed.

Preoperative neck ultrasound (US) for lateral cervical lymph nodes is recommended for all patients undergoing thyroidectomy for thyroid malignancy, but it is operator dependent. We aimed to develop a radiomics signature using US images of the primary tumor to preoperatively predict lateral lymph node metastasis (LNM) in patients with conventional papillary thyroid carcinoma (cPTC). Four hundred consecutive cPTC patients from January 2004 to February 2006 were enrolled as the training cohort, and 368 consecutive cPTC patients from March 2006 to February 2007 served as the validation cohort. A radiomics signature, which consisted of 14 selected features, was generated by the least absolute shrinkage and selection operator (LASSO) regression model in the training cohort. The discriminating performance of the radiomics signature was assessed in the validation cohort with the area under the receiver operating characteristic curve (AUC). The radiomics signature was significantly associated with lateral cervical lymph node status (p < 0.001). The AUC of its performance in discriminating metastatic and non-metastatic lateral cervical lymph nodes was 0.710 (95% CI: 0.649–0.770) in the training cohort and was 0.621 (95% CI: 0.560–0.682) in the validation cohort. The present study showed that US radiomic features of the primary tumor were associated with lateral cervical lymph node status. Although their discriminatory performance was slightly lower in the validation cohort, our study shows that US radiomic features of the primary tumor alone have the potential to predict lateral LNM.

Patients with papillary thyroid carcinoma (PTC) would benefit from risk stratification tools that can aid in planning personalized treatment and follow-up. The aim of this study was to develop a conventional ultrasound (US)-based radiomics signature to estimate disease-free survival (DFS) in patients with conventional PTC. Imaging features were extracted from the pretreatment US images of 768 patients with conventional PTC who were treated between January 2004 and February 2006. The median follow-up period was 117.3 months, with 85 (11.1%) events. A radiomics signature (Rad-score) was generated by using the least absolute shrinkage and selection operator (LASSO) method in Cox regression. The Rad-score was significantly associated with DFS (hazard ratio [HR], 3.087; P < 0.001), independent of clinicopathologic risk factors. A radiomics model which incorporated the Rad-score demonstrated better performance in the estimation of DFS (C-index: 0.777; 95% confidence interval [CI]: 0.735, 0.829) than the clinicopathologic model (C-index: 0.721; 95% CI: 0.675, 0.780). In conclusion, radiomics features from pretreatment US may be potential imaging biomarkers for risk stratification in patients with conventional PTC.