Enhanced U-Net framework with CSN-II and LDM for tuberculin image segmentation

The Tuberculin Skin Test (TST) is an immunological examination method for pulmonary tuberculosis. It involves the intradermal injection of Purified Protein Derivative (PPD), which forms a characteristic induration. Medical personnel can assess the patient’s condition by measuring the size of this induration, providing a basis for large-scale tuberculosis screening and auxiliary diagnosis. This paper proposes an improved U-Net framework. First, a high-precision pre-segmented image is obtained based on cross-color component normalization (CSN-II). Then, the dataset is expanded through the latent space diffusion model (LDM). Finally, the pre-segmented image is combined with the original image, and the Squeeze-and-Excitation (SE-block) attention mechanism is added to the encoding layer and the decoding layer. Convolutional block attention module (CBAM) is integrated in the skip connection to strengthen the channel attention, thereby improving the network segmentation accuracy. The network demonstrated excellent performance across multiple segmentation evaluation metrics (ACC, Recall, DICE, IOU, Precision), indicating the effectiveness of the U-Net framework in the context of Tuberculin Skin Test analysis. This method lays the foundation for the rapid identification of tuberculin.