Efficient Deep Learning Architecture For Detection And Recognition Of Thyroid Nodules

With the development of preventive medicine,people pay more and more attention to health management.With its ever-increasing incidence,the thyroid nodule is one of the most common nodular tumors in the adult population during daily physical examination.Ultrasonography is the primary and preferred screening method for the clinical diagnosis of thyroid nodules.The diagnosis comprises a fine needle aspiration biopsy(FNAB)and a follow-up treatment.Clinically,doctors typically diagnose thyroid nodules by experience.However,this method could result in an ambiguous diagnosis,thereby causing excessive treatments such as unnecessary biopsy and surgery.With the development of artificial intelligence,deep learning has led to great advances in the field of medical image diagnosis.Specifically,the model is constructed to realize automatic detection of thyroid nodules and benign and malignant classification,which can effectively assist doctors in diagnosis.However,due to the different sizes and different shapes and textures of nodules,building accurate and efficient recognition models is crucial.In this work,a series of processing and experiments based on the collection of ultrasound images of thyroid nodules are proposed.The main process is as follows.First,the ultrasound image data is pre-processed.Second,a deep learning model is constructed to detect the location and benign and malignant of thyroid nodules,and a data set is constructed for model evaluation.Finally,a thyroid nodule label and recognition system is designed and developed.The details are as follows:1)A deep learning architecture,you only look once v3 dense multi receptive fields convolutional neural network(YOLOv3-DMRF),based on YOLOv3 is proposed.It comprises a DMRF-CNN and multi-scale detection layers.In DMRF-CNN,dilated convolution with different dilation rates are integrated to continue passing the edge and the texture features to deeper layers.Two different scale detection layers are deployed to recognize the different sizes of the thyroid nodules.2)Two datasets are used to train and evaluate the DMRF-CNN and YOLOv3-DMRF during the experiments.One dataset includes 699 original ultrasound images of thyroid nodules collected from a local health physical center.Finally,10,485 images are obtained after data augmentation.Another dataset is an open-access dataset that includes ultrasound images of 111 malignant and 41 benign thyroid nodules.Receiver operating characteristic(ROC)curve,average precision(AP)and mean average precision(m AP)are used as the metrics to evaluate DMRF-CNN and YOLOv3-DMRF.Experimental results demonstrate that the proposed YOLOv3-DMRF is efficient for detection and recognition of thyroid nodules for ultrasound images.3)Based on YOLOv3-DMRF,an intelligent detection and recognition system for ultrasound thyroid nodules was implemented.After relevant tests,it has been used in cooperative hospitals and provided strong support for subsequent scientific research data collection.