Machine learning models trained on routine brain magnetic resonance imaging free-text reports can effectively stratify the risk of neurosyphilis, according to findings from a 12-year retrospective analysis published in BMC Medical Informatics and Decision Making. In testing conducted on a held-out validation cohort, all computational models developed solely from extracted imaging text features demonstrated area under the receiver operating characteristic curve values exceeding 0.80.
Developing non-invasive risk stratification tools that utilize routinely gathered clinical data offers an approach to identify patients who warrant prioritized cerebrospinal fluid evaluation.
To explore whether routine imaging documentation could support risk assessment, the researchers investigated whether brain magnetic resonance imaging free-text reports could be systematically transformed into interpretable features. The retrospective investigation evaluated patient records spanning from 2012 through 2024 at a national psychiatric specialty center located in China.
The overall study population encompassed 4,652 patients diagnosed with syphilis. Within this cohort, 319 patients were diagnosed with neurosyphilis, while 4,333 patients were classified as having syphilis without neurosyphilis. Using structured features extracted from the unstructured radiology narratives, the investigators developed and internally validated three distinct machine learning architectures: logistic regression utilizing adaptive LASSO, random forest models, and extreme gradient boosting algorithms.
The algorithms were evaluated both with and without patient demographic variables. In the held-out validation dataset, the MRI-only adaptive-LASSO model achieved an ROC-AUC of 0.843, with a 95% confidence interval ranging from 0.786 to 0.893. This standalone imaging model demonstrated a diagnostic sensitivity of 0.811 and a specificity of 0.781. Incorporating demographic variables alongside MRI-derived features into the adaptive-LASSO pipeline increased the predictive performance, yielding an ROC-AUC of 0.868 with a 95% confidence interval of 0.818 to 0.913.
Further evaluations revealed that multiple structured MRI-report variables remained conditionally associated with neurosyphilis following adjustment for demographic covariates. The study authors concluded that routinely generated brain MRI free-text records contain clinically relevant predictive information that can be successfully harnessed for non-invasive neurosyphilis risk stratification.
Subject of Research: Machine learning for neurosyphilis risk stratification from brain MRI reports
Article Title: Machine learning-based risk stratification for neurosyphilis using MRI free-text reports: a 12-year retrospective analysis in China (2012–2024)
Article References: Guo, X., Cheng, J., Zhang, J., Xia, J.-Y., Wang, S., Zheng, H., Song, Y., Chen, W., Gu, W., Qiu, H., Chen, R., Yuan, T.-F., & Zhang, M. (2026). Machine learning-based risk stratification for neurosyphilis using MRI free-text reports: a 12-year retrospective analysis in China (2012–2024). BMC Medical Informatics and Decision Making. https://doi.org/10.1186/s12911-026-03933-z
Image Credits: AI Generated
DOI: 10.1186/s12911-026-03933-z
Keywords: neurosyphilis, syphilis, machine learning, risk stratification, brain MRI, free-text reports, adaptive LASSO, random forest, extreme gradient boosting, cerebrospinal fluid, retrospective cohort, ROC-AUC, internal validation, radiology reports, psychiatric specialty center, medical informatics
Cite Scienmag News
Teresa Odom. (November 18, 2026). Machine Learning Models Stratify Neurosyphilis Risk Using Routine Brain MRI Reports. Scienmag. https://scienmag.com/machine-learning-models-stratify-neurosyphilis-risk-using-routine-brain-mri-reports/
Teresa Odom. "Machine Learning Models Stratify Neurosyphilis Risk Using Routine Brain MRI Reports." Scienmag, 18 November 2026, https://scienmag.com/machine-learning-models-stratify-neurosyphilis-risk-using-routine-brain-mri-reports/. Accessed 18 November 2026.
Teresa Odom. "Machine Learning Models Stratify Neurosyphilis Risk Using Routine Brain MRI Reports." Scienmag. November 18, 2026. https://scienmag.com/machine-learning-models-stratify-neurosyphilis-risk-using-routine-brain-mri-reports/








