Automated and reusable deep learning (AutoRDL) framework for predicting response to neoadjuvant chemotherapy and axillary lymph node metastasis in breast cancer using ultrasound images: a retrospective, multicentre study.

Background: Previous deep learning models have been proposed to predict the pathological complete response (pCR) and axillary lymph node metastasis (ALNM) in breast cancer. Yet, the models often leveraged multiple frameworks, required manual annotation, and discarded low-quality images. We aimed to develop an automated and reusable deep learning (AutoRDL) framework for tumor detection and prediction of pCR and ALNM using ultrasound images with diverse qualities. Methods: The AutoRDL framework includes a You Only Look Once version 5 (YOLOv5) network for tumor detection and a progressive multi-granularity (PMG) network for pCR and ALNM prediction. The training cohort and the internal validation cohort were recruited from Guangdong Provincial People's Hospital (GPPH) between November 2012 and May 2021. The two external validation cohorts were recruited from the First Affiliated Hospital of Kunming Medical University (KMUH), between January 2016 and December 2019, and Shunde Hospital of Southern Medical University (SHSMU) between January 2014 and July 2015. Prior to model training, super-resolution via iterative refinement (SR3) was employed to improve the spatial resolution of low-quality images from the KMUH. We developed three models for predicting pCR and ALNM: a clinical model using multivariable logistic regression analysis, an image model utilizing the PMG network, and a combined model that integrates both clinical and image data using the PMG network. Findings: The YOLOv5 network demonstrated excellent accuracy in tumor detection, achieving average precisions of 0.880-0.921 during validation. In terms of pCR prediction, the combined modelpost-SR3 outperformed the combined modelpre-SR3, image modelpost-SR3, image modelpre-SR3, and clinical model (AUC: 0.833 vs 0.822 vs 0.806 vs 0.790 vs 0.712, all p < 0.05) in the external validation cohort (KMUH). Consistently, the combined modelpost-SR3 exhibited the highest accuracy in ALNM prediction, surpassing the combined modelpre-SR3, image modelpost-SR3, image modelpre-SR3, and clinical model (AUC: 0.825 vs 0.806 vs 0.802 vs 0.787 vs 0.703, all p < 0.05) in the external validation cohort 1 (KMUH). In the external validation cohort 2 (SHSMU), the combined model also showed superiority over the clinical and image models (0.819 vs 0.712 vs 0.806, both p < 0.05). Interpretation: Our proposed AutoRDL framework is feasible in automatically predicting pCR and ALNM in real-world settings, which has the potential to assist clinicians in optimizing individualized treatment options for patients. Funding: National Key Research and Development Program of China (2023YFF1204600); National Natural Science Foundation of China (82227802, 82302306); Clinical Frontier Technology Program of the First Affiliated Hospital of Jinan University, China (JNU1AF-CFTP-2022-a01201); Science and Technology Projects in Guangzhou (202201020022, 2023A03J1036, 2023A03J1038); Science and Technology Youth Talent Nurturing Program of Jinan University (21623209); and Postdoctoral Science Foundation of China (2022M721349).

The added value of apparent diffusion coefficient and microcalcifications to the Kaiser score in the evaluation of BI-RADS 4 lesions.

PURPOSE: To explore the added value of combining microcalcifications or apparent diffusion coefficient (ADC) with the Kaiser score (KS) for diagnosing BI-RADS 4 lesions. METHODS: This retrospective study included 194 consecutive patients with 201 histologically verified BI-RADS 4 lesions. Two radiologists assigned the KS value to each lesion. Adding microcalcifications, ADC, or both these criteria to the KS yielded KS1, KS2, and KS3, respectively. The potential of all four scores to avoid unnecessary biopsies was assessed using the sensitivity and specificity. Diagnostic performance was evaluated by the area under the curve (AUC) and compared between KS and KS1. RESULTS: The sensitivity of KS, KS1, KS2, and KS3 ranged from 77.1% to 100.0%.KS1 yielded significantly higher sensitivity than other methods (P < 0.05), except for KS3 (P > 0.05), most of all, when assessing NME lesions. For mass lesions, the sensitivity of these four scores was comparable (p > 0.05). The specificity of KS, KS1, KS2, and KS3 ranged from 56.0% to 69.4%, with no statistically significant differences(P > 0.05), except between KS1 and KS2 (p < 0.05).The AUC of KS1 (0.877) was significantly higher than that of KS (0.837; P = 0.0005), particularly for assessing NME (0.847 vs 0.713; P < 0.0001). CONCLUSION: KS can stratify BI-RADS 4 lesions to avoid unnecessary biopsies. Adding microcalcifications, but not adding ADC, as an adjunct to KS improves diagnostic performance, particularly for NME lesions. ADC provides no additional diagnostic benefit to KS. Thus, only combining microcalcifications with KS is most conducive to clinical practice.

Comparative Study of C-TIRADS, ACR-TIRADS, and EU-TIRADS for Diagnosis and Management of Thyroid Nodules.

RATIONALE AND OBJECTIVES: Chinese Thyroid Imaging Reporting and Data Systems (C-TIRADS) was developed to provide a more simplified tool for stratifying thyroid nodules. Here we aimed to validate the efficacy of C-TIRADS in distinguishing benign from malignant and in guiding fine-needle aspiration biopsies in comparison with the American College of Radiology TIRADS (ACR-TIRADS) and European TIRADS (EU-TIRADS). MATERIALS AND METHODS: This study retrospectively included 3438 thyroid nodules (≥10 mm) in 3013 patients (mean age, 47.1 years ± 12.9) diagnosed between January 2013 and November 2019. Ultrasound features of the nodules were evaluated and categorized according to the lexicons of the three TIRADS. We compared these TIRADS by using the area under the receiver operating characteristic curve (AUROC), the area under the precision-recall curve (AUPRC), sensitivity, specificity, net reclassification improvement (NRI), and unnecessary fine-needle aspiration biopsy (FNAB) rate. RESULTS: Of the 3438 thyroid nodules, 707 (20.6%) were malignant. C-TIRADS showed higher discrimination performance (AUROC, 0.857; AUPRC, 0.605) than ACR-TIRADS (AUROC, 0.844; AUPRC, 0.567) and EU-TIRADS (AUROC, 0.802; AUPRC, 0.455). The sensitivity of C-TIRADS (85.3%) was lower than that of ACR-TIRADS (89.1%) but higher than that of EU-TIRADS (78.4%). The specificity of C-TIRADS (76.9%) was similar to that of EU-TIRADS (78.9%) and higher than that of ACR-TIRADS (69.5%). The unnecessary FNAB rate was lowest with C-TIRADS (21.2%), followed by ACR-TIRADS (41.7%) and EU-TIRADS (58.3%). C-TIRADS obtained significant NRI for recommending FNAB over ACR-TIRADS (19.0%, P < 0.001) and EU-TIRADS (25.5%, P < 0.001). CONCLUSION: C-TIRADS may be a clinically applicable tool to manage thyroid nodules, which warrants thorough tests in other geographic settings.

Thy-Wise: An interpretable machine learning model for the evaluation of thyroid nodules.

Current risk stratification systems for thyroid nodules suffer from low specificity and high biopsy rates. Recently, machine learning (ML) is introduced to assist thyroid nodule diagnosis but lacks interpretability. Here, we developed and validated ML models on 3965 thyroid nodules, as compared to the American College of Radiology Thyroid Imaging, Reporting and Data System (ACR TI-RADS). Subsequently, a SHapley Additive exPlanation (SHAP) algorithm was leveraged to interpret the results of the best-performing ML model. Clinical characteristics including thyroid-function tests were collected from medical records. Five ACR TI-RADS ultrasonography (US) categories plus nodule size were assessed by experienced radiologists. Random forest (RF), support vector machine (SVM) and extreme gradient boosting (XGBoost) were used to build US-only and US-clinical ML models. The ML models and ACR TI-RADS were compared in terms of diagnostic performance and unnecessary biopsy rate. Among the ML models, the US-only RF model (hereafter, Thy-Wise) achieved the optimal performance. Compared to ACR TI-RADS, Thy-Wise showed higher accuracy (82.4% vs 74.8% for the internal validation; 82.1% vs 73.4% for external validation) and specificity (78.7% vs 68.3% for internal validation; 78.5% vs 66.9% for external validation) while maintaining sensitivity (91.7% vs 91.2% for internal validation; 91.9% vs 91.1% for external validation), as well as reduced unnecessary biopsies (15.3% vs 32.3% for internal validation; 15.7% vs 47.3% for external validation). The SHAP-based interpretation of Thy-Wise enables clinicians to better understand the reasoning behind the diagnosis, which may facilitate the clinical translation of this model.

Intravoxel Incoherent Motion Magnetic Resonance Imaging for Prediction of Induction Chemotherapy Response in Locally Advanced Hypopharyngeal Carcinoma: Comparison With Model-Free Dynamic Contrast-Enhanced Magnetic Resonance Imaging.

BACKGROUND: Multiparametric intravoxel incoherent motion (IVIM) provides diffusion and perfusion information for the treatment prediction of cancer. However, the superiority of IVIM over dynamic contrast-enhanced (DCE) magnetic resonance imaging (MRI) in locally advanced hypopharyngeal carcinoma (LAHC) remains unclear. PURPOSE: To compare the diagnostic performance of IVIM and model-free DCE in assessing induction chemotherapy (IC) response in patients with LAHC. STUDY TYPE: Prospective. POPULATION: Forty-two patients with LAHC. FIELD STRENGTH/SEQUENCE: 3.0 T MRI, including IVIM (12 b values, 0-800 seconds/mm2 ) with a single-shot echo planar imaging sequence and DCE-MRI with a volumetric interpolated breath-hold examination sequence. IVIM MRI is a commercially available sequence and software for calculation and analysis from vendor. ASSESSMENT: The IVIM-derived parameters (diffusion coefficient [D], pseudodiffusion coefficient [D*], and perfusion fraction [f]) and DCE-derived model-free parameters (Wash-in, time to maximum enhancement [Tmax], maximum enhancement [Emax], area under enhancement curve [AUC] over 60 seconds [AUC60 ], and whole area under enhancement curve [AUCw ]) were measured. At the end of IC, patients with complete or partial response were classified as responders according to the Response Evaluation Criteria in Solid Tumors. STATISTICAL TESTS: The differences of parameters between responders and nonresponders were assessed using Mann-Whitney U tests. The performance of parameters for predicting IC response was evaluated by the receiver operating characteristic curves. RESULTS: Twenty-three (54.8%) patients were classified as responders. Compared with nonresponders, the perfusion parameters D*, f, f × D*, and AUCw were significantly higher whereas Wash-in was lower in responders (all P-values <0.05). The f × D* outperformed other parameters, with an AUC of 0.84 (95% confidence interval [CI]: 0.69-0.93), sensitivity of 79.0% (95% CI: 54.4-93.9), and specificity of 82.6% (95% CI: 61.2-95.0). DATA CONCLUSION: The IVIM MRI technique may noninvasively help predict the IC response before treatment in patients with LAHC. LEVEL OF EVIDENCE: 2 TECHNICAL EFFICACY: Stage 2.

A Nomogram for Pretreatment Prediction of Response to Induction Chemotherapy in Locally Advanced Hypopharyngeal Carcinoma.

BACKGROUND: Induction chemotherapy (IC) significantly improves the rate of larynx preservation; however, some patients could not benefit from it. Hence, it is of clinical importance to predict the response to IC to determine the necessity of IC. We aimed to develop a clinical nomogram for predicting the treatment response to IC in locally advanced hypopharyngeal carcinoma. METHODS: We retrospectively include a total of 127 patients with locally advanced hypopharyngeal carcinoma who underwent MRI scans prior to IC between January 2014 and December 2017. The clinical characteristics were collected, which included age, sex, tumor location, invading sites, histological grades, T-stage, N-stage, overall stage, size of the largest lymph node, neutrophil-to-lymphocyte ratio, hemoglobin concentration, and platelet count. Univariate and multivariate logistic regression was used to select the significant predictors of IC response. A nomogram was built based on the results of stepwise logistic regression analysis. The predictive performance and clinical usefulness of the nomogram were determined based on the area under the curve (AUC), calibration curve, and decision curve. RESULTS: Age, T-stage, hemoglobin, and platelet were four independent predictors of IC treatment response, which were incorporated into the nomogram. The AUC of the nomogram was 0.860 (95% confidence interval [CI]: 0.780-0.940), which was validated using 3-fold cross-validation (AUC, 0.864; 95% CI: 0.755-0.973). The calibration curve demonstrated good consistency between the prediction by the nomogram and actual observation. Decision curve analysis shows that the nomogram was clinically useful. CONCLUSION: The proposed nomogram resulted in an accurate prediction of the efficacy of IC for patients with locally advanced hypopharyngeal carcinoma.

Comparison between linear and nonlinear machine-learning algorithms for the classification of thyroid nodules.

BACKGROUND: A key challenge in thyroid carcinoma is preoperatively diagnosing malignant thyroid nodules. The purpose of this study was to compare the classification performance of linear and nonlinear machine-learning algorithms for the evaluation of thyroid nodules using pathological reports as reference standard. METHODS: Ethical approval was obtained for this retrospective analysis, and the informed consent requirement was waived. A total of 1179 thyroid nodules (training cohort, n = 700; validation cohort, n = 479) were confirmed by pathological reports or fine-needle aspiration (FNA) biopsy. The following ultrasonography (US) featu res were measured for each nodule: size (maximum diameter), margins, shape, aspect ratio, capsule, hypoechoic halo, composition, echogenicity, calcification pattern, vascularity, and cervical lymph node status. We analyzed five nonlinear and three linear machine-learning algorithms. The diagnostic performance of each algorithm was compared by using the area under the curve (AUC) of the receiver operating characteristic curve. We repeated this process 1000 times to obtain the mean AUC and 95% confidence interval (CI). RESULTS: Overall, nonlinear machine-learning algorithms demonstrated similar AUCs compared with linear algorithms. The Random Forest and Kernel Support Vector Machines algorithms achieved slightly greater AUCs in the validation cohort (0.954, 95% CI: 0.939-0.969; 0.954 95%CI: 0.939-0.969, respectively) than other algorithms. CONCLUSIONS: Overall, nonlinear machine-learning algorithms share similar performance compared with linear algorithms for the evaluation the malignancy risk of thyroid nodules.

A nomogram for individual prediction of vascular invasion in primary breast cancer.

OBJECTIVES: To explore the feasibility of preoperative prediction of vascular invasion (VI) in breast cancer patients using nomogram based on multiparametric MRI and pathological reports. METHODS: We retrospectively collected 200 patients with confirmed breast cancer between January 2016 and January 2018. All patients underwent MRI examinations before the surgery. VI was identified by postoperative pathology. The 200 patients were randomly divided into training (n = 100) and validation datasets (n = 100) at a ratio of 1:1. Least absolute shrinkage and selection operator (LASSO) regression was used to select predictors most associated with VI of breast cancer. A nomogram was constructed to calculate the area under the curve (AUC) of receiver operating characteristics, sensitivity, specificity, accuracy, positive prediction value (PPV) and negative prediction value (NPV). We bootstrapped the data for 2000 times without setting the random seed to obtain corrected results. RESULTS: VI was observed in 79 patients (39.5%). LASSO selected 10 predictors associated with VI. In the training dataset, the AUC for nomogram was 0.94 (95% confidence interval [CI]: 0.89-0.99, the sensitivity was 78.9% (95%CI: 72.4%-89.1%), the specificity was 95.3% (95%CI: 89.1%-100.0%), the accuracy was 86.0% (95%CI: 82.0%-92.0%), the PPV was 95.7% (95%CI: 90.0%-100.0%), and the NPV was 77.4% (95%CI: 67.8%-87.0%). In the validation dataset, the AUC for nomogram was 0.89 (95%CI: 0.83-0.95), the sensitivity was 70.3% (95%CI: 60.7%-79.2%), the specificity was 88.9% (95%CI: 80.0%-97.1%), the accuracy was 77.0% (95%CI: 70.0%-83.0%), the PPV was 91.8% (95%CI: 85.3%-98.0%), and the NPV was 62.7% (95%CI: 51.7%-74.0%). The nomogram calibration curve shows good agreement between the predicted probability and the actual probability. CONCLUSION: The proposed nomogram could be used to predict VI in breast cancer patients, which was helpful for clinical decision-making.

Development and validation of an ultrasound-based nomogram to improve the diagnostic accuracy for malignant thyroid nodules.

OBJECTIVES: The aim of this study was to develop an ultrasound-based nomogram to improve the diagnostic accuracy of the identification of malignant thyroid nodules. METHODS: A total of 1675 histologically proven thyroid nodules (1169 benign, 506 malignant) were included in this study. The nodules were grouped into the training dataset (n = 700), internal validation dataset (n = 479), or external validation dataset (n = 496). The grayscale ultrasound features included the nodule size, shape, aspect ratio, echogenicity, margins, and calcification pattern. We applied least absolute shrinkage and selection operator (lasso) regression to select the strongest features for the nomogram. Nomogram discrimination (area under the receiver operating characteristic curve, AUC) and calibration were assessed. The nomogram was subjected to bootstrapping validation (1000 bootstrap resamples) to calculate a mean AUC and 95% confidence interval (CI). RESULTS: The nomogram showed good discrimination in the training dataset, with an AUC of 0.936 (95% CI: 0.918-0.953) and good calibration. Application of the nomogram to the internal validation dataset also resulted in good discrimination (AUC: 0.935; 95% CI, 0.915-0.954) and good calibration. The model tested in an external validation dataset demonstrated a lower AUC of 0.782 (95% CI: 0.776-0.789). CONCLUSIONS: This ultrasound-based nomogram can be used to quantify the probability of malignant thyroid nodules. KEY POINTS: • Ultrasound examination is helpful in the differential diagnosis of malignant and benign thyroid nodules. • However, ultrasound accuracy relies heavily on examiner experience. • A less subjective diagnostic model is desired, and the developed nomogram for thyroid nodules showed good discrimination and good calibration.

Development of a preprocedure nomogram for predicting contrast-induced acute kidney injury after coronary angiography or percutaneous coronary intervention.

Most of the risk models for predicting contrast-induced acute kidney injury (CI-AKI) are available for postcontrast exposure prediction, thus have limited values in practice. We aimed to develop a novel nomogram based on preprocedural features for early prediction of CI-AKI in patients after coronary angiography (CAG) or percutaneous coronary intervention (PCI). A total of 245 patients were retrospectively reviewed from January 2015 to January 2017. Least absolute shrinkage and selection operator (Lasso) regression model was applied to select most strong predictors for CI-AKI. The CI-AKI risk score was calculated for each patient as a linear combination of selected predictors that were weighted by their respective coefficients. The discrimination of nomogram was assessed by C-statistic. The occurrence of CI-AKI was 13.9% (34 out of 245). We identified ten predictors including sex, diabetes mellitus, lactate dehydrogenase level, C-reactive protein, years since drinking, chronic kidney disease (CKD), stage of CKD, stroke, acute myocardial infarction, and systolic blood pressure. The CI-AKI prediction nomogram obtained good discrimination (C-statistic, 0.718, 95%CI: 0.637-0.800, p = 7.23 × 10-5). The cutoff value of CI-AKI risk score was -1.953. Accordingly, the novel nomogram we developed is a simple and accurate tool for preprocedural prediction of CI-AKI in patients undergoing CAG or PCI.