Ensemble-based Deep Learning Framework for Tuberculosis Detection in Radiographs

Tuberculosis (TB) continues to be a serious worldwide health concern, and in order to effectively treat it, it is necessary to diagnose it accurately and in a timely manner. The purpose of this research is to provide a robust framework for tuberculosis diagnosis that makes use of a deep learning-based model called DenseNet201 in conjunction with a stacking ensemble method in order to improve diagnostic accuracy from chest X-ray pictures. Deep features are extracted in an effective manner by the DenseNet201 architecture, which captures intricate patterns that are suggestive of tuberculosis. Support Vector Machine (SVM), Random Forest (RF), and XGBoost are the three base classifiers that receive these features as input. The outputs of these classifiers are then aggregated with a meta-classifier to get the final prediction. This helps to further improve the performance of the prediction. This stacking strategy helps to minimize the flaws of individual classifiers, hence minimizing the number of false positives and negatives seen in medical diagnostics, which is an extremely important skill. A high accuracy of 95.3% and an area under the curve (AUC) of 0.973 are demonstrated by the model after it has been trained and assessed on the TBX11K dataset. This greatly outperforms the performance of older approaches. In order to increase the quality of the data and better the performance of the model, preprocessing processes were used. These stages included normalization, data augmentation, and contrast enhancement. Due to the fact that the suggested ensemble model generalizes well to data that has not been encountered before, it is a trustworthy and effective instrument for early and accurate tuberculosis screening. This method has the potential to be implemented in clinical settings that are representative of the real world in order to provide assistance to medical professionals in the diagnosis of tuberculosis.