Integrated deep learning multi-label identification method based on TI-RADS

Abstract

The invention relates to an integrated deep learning multi-label recognition method based on TI-RADS, and the method comprises the following steps: S1, preprocessing: carrying out the preprocessing ofa collected original thyroid ultrasound image, and the preprocessing comprises the segmentation of a nodule boundary of the original thyroid ultrasound image and the extraction of a nodule region ofinterest; S2, performing feature engineering, wherein the feature engineering is used for extracting geometrical features and texture features of the original thyroid ultrasonic image preprocessed inthe step S1; S3, building a model: performing feature fusion on the Effective Net model, the feature engineering and the FPN network model through a Concatenate function to obtain a deep learning model; and S4, inputting the original thyroid ultrasound image preprocessed in the step S1 and the geometrical characteristics and texture characteristics extracted in the step S2 into the deep learning model in the step S3, and outputting a multi-label classification result. The method has the advantages of being high in releasability and accurate in classification result.

Description

technical field [0001] The invention belongs to the technical field of auxiliary diagnosis methods for thyroid nodules, and is an integrated deep learning multi-label recognition method based on TI-RADS. Background technique [0002] Thyroid nodules are lumps growing in the thyroid gland. In recent years, the incidence has been on the rise. Most nodules are benign, but 5%-15% are malignant. Ultrasonography has become the preferred method for doctors to diagnose thyroid nodules due to its low cost, safety and non-invasive advantages. Clinically, doctors usually diagnose thyroid nodules based on experience. They observe ultrasound images and compare them with scoring standards to obtain a risk grade score for thyroid nodules. The American College of Radiology (ACR) proposed the Thyroid Imaging Reporting and Data System (TI-RADS), which analyzed five characteristics of thyroid nodules: composition, echo, shape, edge, and hyperechoic. Manually diagnosing thyroid ultrasound ima...

AI Technical Summary

Problems solved by technology

[0004] However, the existing technology usually uses machine learning or deep learning methods to extract features and classify them. Some existing machine learning methods often classify a single feature of a certain class, and the accuracy is often not ideal. somewhat lacking in interpretability

Images