Research On Multi-Label Classification Based On Pathological Features Of Thyroid Nodules

With the increasing pressure in people’s modern life,the incidence of thyroid nodules has increased year by year.Some malignant nodules can cause thyroid cancer.Early and accurate identification of thyroid nodules is of great significance to clinical diagnosis and treatment.Ultrasonography is the most commonly used method to examine thyroid nodules.Due to the low contrast of ultrasound images,unclear edge of the nodules,and other factors,artificial recognition is inefficient,with low accuracy,and prone to over-treatment.Artificial intelligence has played a great role in assisting medical intelligent diagnosis,and it also has a good effect in distinguishing benign and malignant thyroid nodules.However,most of the researches focus on the feature extraction of ultrasound images,ignoring the pathological features of the nodules themselves,which makes the recognition accuracy and interpretability problems.Based on the clinical diagnosis of ultrasound doctors,this paper proposes an integrated deep learning multi-label feature recognition model based on the ACR TI-RADS standard.This model integrates the extraction of pathological features of thyroid nodules,the experiments have proved that this model can be effectively used for the identification of benign and malignant thyroid nodules,thereby assisting doctors in clinical diagnosis.The main research contents of this paper include:1)Acquire ultrasound images of thyroid nodules,first preprocess the image data,segment the boundaries of the nodules,extract regions of interest,establish feature engineering,extract the geometric and texture features of the nodules,and denoise the ultrasound images;2)An integrated deep learning multi-label classification model is proposed,which is mainly composed of Efficient Net,feature engineering,and feature pyramid network.It is based on the five features of composition,echogenicity,shape,margin,and echogenic foci proposed by the ACR TI-RADS standard,and combines traditional feature extraction and deep learning feature extraction.The two groups of experimental datasets come from the data collected by the local provincial-level tertiary hospitals and the MICCAI 2020 public competition dataset.The accuracy,ROC curve and AUC value indicators are used to evaluate the model.The experimental results show that the model proposed in this paper is effective for multi-label feature recognition of thyroid nodules;3)Furthermore,the integrated deep learning multi-label classification model based on thyroid nodules proposed in this paper is used to classify benign and malignant thyroid nodules,and related experiments are carried out.First,calculate the total score based on the ACR TI-RADS scoring standard on the MICCAI 2020 public competition dataset to judge benign and malignant,and compare it with the real label of biopsy puncture surgery.Compare and analyze the integrated deep learning multilabel classification model proposed in this article with other models.The experimental results show that the model proposed in this paper has good classification results for the classification of benign and malignant thyroid nodules.