Tuberculosis identification and detection application using deep learning—cloud based web

Mycobacterium tuberculosis is the bacteria that causes tuberculosis (TB), a communicable disease that mostly affects the lungs but can also affect other regions of body. The World Health Organization (WHO) forecasts that approximately 10 million individuals worldwide contracted tuberculosis (TB) in 2019. Tuberculosis is a serious global health concern. Effective disease management and transmission prevention of tuberculosis (TB) depend on early detection and treatment. Through the use of image pre-processing, picture segmentation, deep learning classification, data augmentation methods, we have successfully identified tuberculosis from chest X-ray images in this work. For this investigation, a database of 4,000 TB-infected and 4,000 normal chest X-ray pictures was assembled utilizing number of public databases. For transfer learning from their pre-trained initial weights, nine deep CNNs and SVM (ResNet18, ResNet50, ResNet101, Vgg19, ChexNet, DenseNet201, SqueezeNet, InceptionV3, and MobileNet) were employed. They underwent testing, validation, and training to identify TB and non-TB normal cases. This work involved three separate experiments: segmenting X-ray images utilizing 2 distinct U-net methods, classifying X-ray images, and segmenting lung images. The specificity and sensitivity of conventional TB diagnostic procedures, such as sputum microscopy and culture-based approaches, are limited, especially when it comes to early detection of TB. Support Vector Machine (SVM), one of machine learning methods, has demonstrated promise in accurately detecting tuberculosis. However, segmented lung image classification fared better than whole X-ray image classification; for the segmented lung image classification, DenseNet201 performed better in terms of accuracy, precision, sensitivity, F1-score, and specificity. Additionally, a visualization technique was employed in the article to validate that CNN learns primarily from segmented lung areas, leading to increased detection accuracy. In general, a cloud-based online application for SVM-based speedy and accurate tuberculosis detection would be a useful weapon in fight against this illness, empowering medical professionals to make better decisions regarding diagnosis and treatment.