Optimizing Tuberculosis Detection: A Modified DenseNet121 Approach for High-Accuracy Chest X-Ray Analysis

This research study presents a DenseNet 121 modified for detecting TB from pictures of the chest X-rays. Binary classification tuning was made using 700 Tuberculosis-positive photos available in the Tuberculosis Chest X-ray Database, 7000 Tuberculosis-positive more available in NIAID TB portal, and 3500 normal images. DenseNet121 design has been originally praised for its effectiveness in feature extraction It was decided to transform the DenseNet121 design with a global average pooling layer and sigmoid activation. When control was made between normal status and TB-positive, the accuracy of the model was checked to be approximately 90% correct. Ambiguity characteristics for normal pictures gave a precision of 0.90 and recall of 1.00 suggesting very excellent accuracy for non-TB diagnosis. While the precise rate was 1.00, the rate for cases positive for tuberculosis was only 0.43. Further promoted the discriminative ability of the model, and by the ROC curve analysis, the AUC of 0.976 obtained showed the efficiency of the model. These results may necessitate further attempts in I-FIND to confirm the viability of the DenseNet121 model for discerning TB from non-TB and increasing TB POS case finding.