Deep learning on longitudinal CT scans: automated prediction of treatment outcomes in hospitalized tuberculosis patients

Development of a clinical prediction model for poor treatment outcomes in the intensive phase in patients with initial treatment of pulmonary tuberculosis

Background

A prediction model is hereby developed to identify poor treatment outcomes during the intensive phase in patients with initial treatment of pulmonary tuberculosis (TB).

Methods

The data of inpatients with pulmonary TB were collected from a tertiary hospital located in Southeastern China from July 2019 to December 2023. The included patients were divided into the modeling group and the validation group. The outcome indicator was based on a comparison of pulmonary CT findings before and after the two-month intensive phase of anti-TB treatment. In the modeling group, the independent risk factors of pulmonary TB patients were obtained through logistic regression analysis and then a prediction model was established. The discriminative ability (the area under the curve of the receiver operating characteristic, AUC), its calibration (GiViTI calibration chart), and its clinical applicability (decision curve analysis, DCA) were respectively evaluated. In addition, the prediction effectiveness was compared with that of the machine learning model.

Results

A total of 1,625 patients were included in this study, and 343 patients had poor treatment outcomes in the intensive phase of anti-TB treatment. Logistic regression analysis identified several independent risk factors for poor treatment outcomes, including diabetes, cavities in the lungs, tracheobronchial TB, increased C-reactive protein, and decreased hemoglobin. The AUC values were 0.815 for the modeling group and 0.851 for the validation group. In the machine learning models, the AUC values of the random forest model and the integrated model were 0.821 and 0.835, respectively.

Conclusion

The prediction model established in this study presents good performance in predicting poor treatment outcomes during the intensive phase in patients with pulmonary TB.

Prediction of tuberculosis treatment outcomes using biochemical makers with machine learning

BACKGROUND

Tuberculosis (TB) continues to pose a significant threat to global public health. Enhancing patient prognosis is essential for alleviating the disease burden.

OBJECTIVE

This study aims to evaluate TB prognosis by incorporating treatment discontinuation into the assessment framework, expanding beyond mortality and drug resistance.

METHODS

Seven feature selection methods and twelve machine learning algorithms were utilized to analyze admission test data from TB patients, identifying predictive features and building prognostic models. SHapley Additive exPlanations (SHAP) were applied to evaluate feature importance in top-performing models.

RESULTS

Analysis of 1,086 TB cases showed that a K-Nearest Neighbor classifier with Mutual Information feature selection achieved an area under the receiver operation curve (AUC) of 0.87 (95% CI: 0.83-0.92). Key predictors of treatment failure included elevated levels of 5'-nucleotidase, uric acid, globulin, creatinine, cystatin C, and aspartate transaminase. SHAP analysis highlighted 5'-nucleotidase, uric acid, and globulin as having the most significant influence on predicting treatment discontinuation.

CONCLUSION

Our model provides valuable insights into TB outcomes based on initial patient tests, potentially guiding prevention and control strategies. Elevated biomarker levels before therapy are associated with increased risk of treatment discontinuation, indicating their potential as early warning indicators.

The impact of maximum cross-sectional area of lesion on predicting the early therapeutic response of multidrug-resistant tuberculosis

BACKGROUND

Early evaluation of culture conversion after 6-month treatment of multidrug-resistant tuberculosis (MDR-TB) is vital for outcome prediction. This study aims to merge the maximum lesion cross-sectional area observed via computed tomography (CT) imaging during treatment to predict therapeutic response.

METHODS

We retrospectively involved MDR-TB patients who completed 6 months of treatment from two hospitals. Patients were categorized into culture conversation and no culture conversation groups based on sputum culture results. The data from the two hospitals were used as internal training and external testing cohorts, respectively. Logistic regression and random forest models were developed using the maximum lesion cross-sectional area and most important predictive features. The model performance was evaluated using the area under the curve (AUC), accuracy, sensitivity, specificity, and F1 score.

RESULTS

In the model without the maximum lesion cross-sectional area to predict culture conversion for MDR-TB after 6 months of treatment, logistic regression and random forest models achieved AUC values of 0.796 and 0.958, sensitivities of 0.725 and 0.993, and F1 scores of 0.803 and 0.957 in the training cohort, respectively. In the testing cohort, logistic regression and random forest models achieved AUC values of 0.889 and 0.855, respectively. Evaluating the maximum lesion cross-sectional area at baseline, 2 months, and 6 months, logistic regression and random forest models in the training cohort yielded AUC values of 0.819 and 0.998, sensitivities of 0.674 and 1.000, and F1 scores of 0.772 and 0.986. In the testing cohort, the AUC values were 0.869 and 0.920, sensitivities were 0.933 and 1.000, and F1 scores were 0.848 and 0.841, respectively.

CONCLUSIONS

The integration of maximum lesion cross-sectional area during treatment can improve the prediction of early treatment response in MDR-TB. When applied in a clinical setting, the random forest model is more suitable for guiding appropriate treatment plans quickly.

Clinical utilization of artificial intelligence in predicting therapeutic efficacy in pulmonary tuberculosis

Traditional methods for monitoring pulmonary tuberculosis (PTB) treatment efficacy lack sensitivity, prompting the exploration of artificial intelligence (AI) to enhance monitoring. This review investigates the application of AI in monitoring anti-tuberculosis (ATTB) treatment, revealing its potential in predicting treatment duration, adverse reactions, outcomes, and drug resistance. It provides important insights into the potential of AI technology to enhance monitoring and management of ATTB treatment. Systematic search across six databases from 2013 to 2023 explored AI in forecasting PTB treatment efficacy. Support vector machine and convolutional neural network excel in treatment duration prediction, while random forest, artificial neural network, and classification and regression tree show promise in forecasting adverse reactions and outcomes. Neural networks and random forest are effective in predicting drug resistance. AI advancements offer improved monitoring strategies, better patient prognosis, and pave the way for future AI research in PTB treatment monitoring.