Research On Liver Tumor Detection Method Based On Deep Learnin

Diagnostic imaging,with its intuitive imaging characteristics,has gradually become one of the main tools for screening lesions in clinical diagnosis,especially in cancer diagnosis has played a high application value.With the rapid development of artificial intelligence technology in the field of computer vision,powerful deep learning algorithms are gradually entering the medical field.This thesis presents a series of studies based on deep learning for lesion detection and lesion segmentation in CT images with complex backgrounds.1.Design of a liver tumour target detection algorithm based on improved YOLOv5.The YOLOv5 network fused with the adaptive attention module is used in CT image liver tumor detection.By introducing an adaptive attention module consisting of both channel attention and spatial attention into the feature prediction structure of YOLOv5,the filtered and weighted feature vectors are replaced by the original feature vectors,so that important tumour target features occupy a greater proportion of the network processing,ultimately enhancing the network’s feature recognition capability for tumour targets in complex backgrounds.The loss function of the detection model is also optimised according to the tumour characteristics in the CT images,thus improving the detection of tumour targets in the CT images.Finally,the effectiveness of the improved algorithm is verified through comparative experiments.Compared with the original YOLOv5 algorithm,the average accuracy of the algorithm in this chapter is improved by 14.5%,and a more excellent prediction frame is obtained in the result prediction.2.Design of a liver tumour segmentation algorithm based on improved UNet++ By introducing a channel adaptive attention module into the downsampling coding layer structure of UNet++,the network’s ability to capture interactive information in the channel domain is improved,and the impact of non-critical information features on network learning is reduced.The network’s ability to aggregate feature information at multiple resolutions is enhanced by the inclusion of a spatially adaptive perceptual field module,which improves the network’s feature extraction performance for small-sized targets.The experimental results show that the improved algorithm improves the segmentation performance of small-sized targets and targets in complex backgrounds regions.Compared to the original UNet++ algorithm,the average accuracy of the segmentation is improved by10.4%.3.Design of liver tumor diagnosis interface based on Py Qt5 The practicality of the algorithm is enhanced by the design and implementation of the human-computer interaction module.The main work content includes: I.Complete porting of all algorithm modules to the tumour diagnosis system interface.Ⅱ.Extending the corresponding auxiliary functions to maximise the efficiency of the use of the algorithms.