Efficient and Fast Tuberculosis Detection in Chest X-rays Using MLP and DCT

In medical image classification, Convolutional Neural Network (CNN) architectures are widely utilized due to their ability to classify complex image patterns. However, training them can be computationally expensive. This paper proposes a lightweight framework that balances high accuracy with reduced computational cost by employing a Multi-Layer Perceptron (MLP) architecture combined with Discrete Cosine Transform (DCT) for the classification of chest X-ray images in tuberculosis detection. A test accuracy of 98.25% is achieved, along with a recall of 97.14%, demonstrating comparable performance to deeper CNN models. This study can be extended to embedded applications for low-power, resource-constrained devices, enabling real-time disease detection on portable medical imaging equipment.