Improved U-Net tuberculosis image segmentation

Tuberculosis is a highly contagious and potentially fatal disease. Early detection and appropriate treatment can significantly improve patients' survival chances. Chest X-rays are widely used as the primary tool for diagnosing tuberculosis due to their convenience and speed. However, the complexity of chest X-ray images and the large amount of data involved may interfere with the diagnostic efficiency and accuracy of doctors. Therefore, the application of computer technology for automatic detection of tuberculosis is particularly crucial, as it can assist doctors in achieving more accurate and rapid diagnoses. This study addresses the high challenge of lung tuberculosis medical image segmentation and proposes a novel network architecture called EfficientDATBUnet. This architecture is based on U-Net, incorporates a pre-trained EfficientNet-B7 to optimize the encoder, and combines a dual aggregation Transformer block (DATB) to significantly enhance the model's feature extraction capabilities. In the decoder part, the ELU activation function is used instead of the traditional ReLU to reduce the phenomenon of neuron death and accelerate the model training process. The results of this study are not only innovative in theory but also of great value in practical applications, providing new technical support for the early diagnosis and treatment of tuberculosis.