Nodule Classification Algorithm For Thyroid Ultrasound Images Based On Deep Learning

Recent years,the combination of medical images and artificial intelligence has become a research hotspot in the digital medical industry.Artificial intelligence technology processing medical images and participating in medical diagnosis has become an important research direction in the field of image processing.Medical image segmentation is one of the most important research hotspots.In this paper,we divide the ultrasound image of thyroid nodules based on the full convolutional neural network.Aiming at the problems of low resolution and too much interference information of ultrasound images,we propose a layer-by-layer semantic segmentation algorithm,which is divided into three levels: the first is the ROI segmentation of thyroid ultrasound images.We use the full convolutional neural network which the up sampling is symmetry with the down sampling to remove the interference of the instrument information and extract the ultrasound reflection region.Secondly,we locate the thyroid nodules roughly in the ROI region,including manual marker detection,localization and locating the nodules according to the artificial markers.This step can remove the ultrasonic interference such as blood vessels around the nodules.Finally,we segment the coarsely located nodule images finely.A deep full convolution neural network structure with VGG19 as the down sampling layers is used to extract the features of thyroid nodule and segment the coarse localization images.The algorithm gradually reduces various interference information in the ultrasound image through the threelevel series steps,and finally greatly improves the segmentation accuracy.In this paper,we use the thyroid ultrasound images provided by Tianjin Medical University Cancer Hospital as the data set,and compare the results of layer-by-layer algorithm with the results of the original ultrasound image without ROI segmentation and coarse position and the ROI image without coarse position.A large number of experiments have proved that the algorithm in this paper has greatly improved the segmentation accuracy of thyroid ultrasound images.Compared with the results of manual labeling,it is proved that the algorithm can provide diagnostic related information,which can effectively help doctors in the diagnosis of benign and malignant nodules.