Pulmonary nodule diagnosis method based on double-mode extreme learning machine

Abstract

The invention discloses a pulmonary nodule diagnosis method based on a double-mode extreme learning machine. The method comprises the steps that firstly, pulmonary nodule PET-CT images are subjected to autonomous characteristic learning by using a self-encoding network with the supervision depth respectively, then, the extracted CT characteristics and PET characteristics are subjected to characteristics layer fusion, and finally, the fusion characteristics are classified by using a classifier. The method is simple and easy to implement, the identification accuracy is high, and according to the PET-CT images of pulmonary nodules, the benign and malignancy of the pulmonary nodules in the PET-CT images can be automatically and accurately identified based on the deep learning technology and the characteristic fusion method respectively.

Description

technical field [0001] The invention relates to the diagnosis of pulmonary nodules, in particular to a method for diagnosing pulmonary nodules based on a dual-mode extreme learning machine. Background technique [0002] In the existing auxiliary diagnosis system for pulmonary nodules, the classification of benign and malignant nodules is mainly based on single modality image features (such as CT, MRI). Relying on a single feature will bring the risk of misdiagnosis. [0003] In the existing self-encoding network, the initial weights are randomly generated and unsupervised learning is adopted, and the resulting image features are not very distinguishable, which is not conducive to classification. [0004] In order to solve the above problems, in the present invention, PET-CT images are used to diagnose pulmonary nodules by combining two modal image features, and the autoencoding algorithm is optimized by using restricted difference and supervised encoding methods in the autoe...

AI Technical Summary

Problems solved by technology

[0002] In the existing aided diagnosis system for pulmonary nodules, the classification of benign and malignant nodules is mainly based on single-modal image features (such as CT, MRI). Relying on a single feature will bring the risk of misdiagnosis

[0003] In the existing self-encoder network, the initial weights are randomly generated and unsupervised learning is adopted, the resulting image features are not very distinguishable, which is not conducive to classification

Images