Medical Imaging Tumor Segmentation Algorithm Based on W-DeepLabV3+ Model

Accurately segmenting tumor areas from medical images helps doctors to locate, evaluate and plan treatments for tumors. Due to the problem of blurred boundaries of tumors in medical images, it brings challenges to the segmentation task. In order to segment the tumor area accurately, a medical imaging tumor segmentation algorithm based on the W-DeepLabV3+ model is proposed. The algorithm first uses a feature extraction network that combines stacking different numbers of convolutional layers and splicing to obtain more tumor feature information. Then a self-attention mechanism is introduced to make the model pay more attention to the tumor area of interest. Finally, a series-parallel feature pyramid is used to enable the network to capture tumor features at different scales more quickly and accurately, improving the accuracy of the segmentation task. The algorithm was trained and tested on the brain tumor data set and the rectal tumor data set, with mIou of 93.87% and 92.67%, mDice of 95.93% and 96.68%, and compared with the existing mainstream semantic segmentation algorithm models MobileNetV2-DeepLabV3+ and InR-Unet (Inverted Residual Unet) conducted comparative experiments, in which the mIou on the brain tumor data set was 2.62% and 2.46% higher. Experimental results show that the proposed algorithm can segment tumor areas more accurately.