Dataset Information

Machine Learning Models to Improve the Differentiation Between Benign and Malignant Breast Lesions on Ultrasound: A Multicenter External Validation Study.

ABSTRACT:

Purpose

This study aimed to establish and evaluate the usefulness of a simple, practical, and easy-to-promote machine learning model based on ultrasound imaging features for diagnosing breast cancer (BC).

Materials and methods

Logistic regression, random forest, extra trees, support vector, multilayer perceptron, and XG Boost models were developed. The modeling data set of 1345 cases was from a tertiary class A hospital in China. The external validation data set of 1965 cases were from 3 tertiary class A hospitals and 2 primary hospitals. The area under the receiver operating characteristic curve (AUC) was used as the main evaluation index, and pathological biopsy was used as the gold standard for evaluating each model. Diagnostic capability was also compared with that of clinicians.

Results

Among the six models, the logistic model showed superior diagnostic efficiency, with an AUC of 0.771 and 0.906 and Brier scores of 0.181 and 0.165 in the test and validation sets, respectively. The AUCs of the clinician diagnosis and the logistic model were 0.913 and 0.906. Their AUCs in the tertiary class A hospitals were 0.915 and 0.915, respectively, and were 0.894 and 0.873 in primary hospitals, respectively.

Conclusion

The externally validated logical model can be used to distinguish between malignant and benign breast lesions in ultrasound images. Compared with clinician diagnosis, the logistic model has better diagnostic efficiency, making it potentially useful to assist in screening, particularly in lower level medical institutions.

Trial registration

http://www.clinicaltrials.gov. ClinicalTrials.gov ID: NCT03080623.

SUBMITTER: Huo L

PROVIDER: S-EPMC8057795 | biostudies-literature |

REPOSITORIES: biostudies-literature

ACCESS DATA

Similar Datasets

Project description:ObjectiveTo identify the diagnostic ability of precontrast and contrast-enhanced ultrasound (CEUS) in differentiating between malignant and benign solid pancreatic lesions (MSPLs and BSPLs) and to develop an easy-to-use diagnostic nomogram.Materials and methodsThis study was approved by the institutional review board. Patients with pathologically confirmed solid pancreatic lesions were enrolled from one tertiary medical centre from March 2011 to June 2021 and in two tertiary institutions between January 2015 and June 2021. A prediction nomogram model was established in the training set by using precontrast US and CEUS imaging features that were independently associated with MSPLs. The performance of the prediction model was further externally validated.ResultsA total of 155 patients (mean age, 55 ± 14.6 years, M/F = 84/71) and 78 patients (mean age, 59 ± 13.4 years, M/F = 36/42) were included in the training and validation cohorts, respectively. In the training set, an ill-defined border and dilated main pancreatic duct on precontrast ultrasound, CEUS patterns of hypoenhancement in both the arterial and venous phases of CEUS, and hyperenhancement/isoenhancement followed by washout were independently associated with MSPLs. The prediction nomogram model developed with the aforementioned variables showed good performance in differentiating MSPLs from BSPLs with an area under the curve (AUC) of 0.938 in the training set and 0.906 in the validation set.ConclusionHypoenhancement in all phases, hyperenhancement/isoenhancement followed by washout on CEUS, an ill-defined border, and a dilated main pancreatic duct were independent risk factors for MSPLs. The nomogram constructed based on these predictors can be used to diagnose MSPLs.Key points• An ill-defined border and dilated main pancreatic duct on precontrast ultrasound, hypoenhancement in all phases of CEUS, and hyperenhancement/isoenhancement followed by washout were independently associated with MSPLs. • The ultrasound-based prediction model showed good performance in differentiating MSPLs from BSPLs with an AUC of 0.938 in the training set and 0.906 in the external validation set. • An ultrasound-based nomogram is an easy-to-use tool to differentiate between MSPLs and BSPLs with high efficacy.

Project description:Background: Because neurofibromatosis type I (NF1) is a cancer predisposition disease, it is important to distinguish between benign and malignant lesions, especially in the craniofacial area. Purpose: The purpose of this study is to improve effectiveness in the diagnostic performance in discriminating malignant from benign craniofacial lesions based on computed tomography (CT) using a Keras-based machine-learning model. Methods: The Keras-based machine learning technique, a neural network package in the Python language, was used to train the diagnostic model on CT datasets. Fifty NF1 patients with benign craniofacial neurofibromas and six NF1 patients with malignant peripheral nerve sheath tumors (MPNSTs) were selected as the training set. Three validation cohorts were used: validation cohort 1 (random selection of 90% of the patients in the training cohort), validation cohort 2 (an independent cohort of 9 NF1 patients with benign craniofacial neurofibromas and 11 NF1 patients with MPNST), and validation cohort 3 (eight NF1 patients with MPNST, not restricted to the craniofacial area). Sensitivity and specificity were tested using validation cohorts 1 and 2, and generalizability was evaluated using validation cohort 3. Results: A total of 59 NF1 patients with benign neurofibroma and 23 NF1 patients with MPNST were included. A Keras-based machine-learning model was successfully established using the training cohort. The accuracy was 96.99 and 100% in validation cohorts 1 and 2, respectively, discriminating NF1-related benign and malignant craniofacial lesions. However, the accuracy of this model was significantly reduced to 51.72% in the identification of MPNSTs in different body regions. Conclusion: The Keras-based machine learning technique showed the potential of robust diagnostic performance in the differentiation of craniofacial MPNSTs and benign neurofibromas in NF1 patients using CT images. However, the model has limited generalizability when applied to other body areas. With more clinical data accumulating in the model, this system may support clinical doctors in the primary screening of true MPNSTs from benign lesions in NF1 patients.