Research on Tumor Images Segmentation based on Improved Convolutional Neural Network

Tumors are one of the most common tumors in humans, and the main treatment at present is surgical removal. Preoperative planning of tumors usually requires a professional doctor to manually segment the images, but this process is limited by the doctor's technical experience, which is time-consuming and labor-intensive. In addition, segmentation is extremely difficult due to the complexity and morphological diversity of tumor lesions. We propose a tumor image segmentation method based on improved Convolutional Neural Network (CNN). The model enhances the feature extraction ability by introducing a deep separable convolutional layer and batch normalization, while using dilated convolution to increase the receptive field, thereby improving the accuracy of segmentation. We also incorporated attention mechanisms into the model to focus more accurately on the tumor area and further optimize the segmentation effect. In order to verify the effectiveness of the model, we conducted experimental analysis on multiple publicly available medical imaging datasets. Experimental results show that the improved convolutional neural network significantly improves the accuracy and efficiency of segmentation compared with traditional methods in tumor image segmentation tasks.