Lung nodule screening method based on deep learning

Abstract

The invention discloses a lung nodule screening method based on deep learning. The method comprises the following steps: S1, preprocessing a lung CT image to extract a lung parenchymal part; S2, detecting the pre-processed lung CT image by a lung nodule detection network, and detecting the location of a lung nodule to obtain the location of a lung nodule candidate area; S3, further classifying thelung nodule candidate area, and scanning a real nodule area by a deep learning classification network; and S4, performing nodule malignant degree grading on the real nodule area by the deep learningclassification network. According to the lung nodule screening method, pulmonary nodule detection is performed on the CT image to distinguish the malignant degree of a target nodule, thereby greatly reducing the workload of doctors and improving the diagnostic efficiency of the doctors.

Description

technical field [0001] The present invention relates to image recognition and medical imaging, in particular to a method for screening lung nodules based on deep learning. Background technique [0002] Lung cancer is the cancer with the highest mortality rate among cancers, which is extremely harmful to human health. With the increasingly serious problems such as environmental pollution, the incidence and mortality of lung cancer are increasing year by year. In 2012, it is estimated that there were 1.8 million new lung cancer patients worldwide, accounting for 13% of all lung cancer cases in that year. In China and other East Asian countries, more than half of lung cancer deaths were caused by atmospheric particles. In addition, the results of the cause of death survey of Chinese residents show that lung cancer is the fastest growing malignant tumor. [0003] Lung cancer is difficult to detect in the early stage, the symptoms are mild, the onset time is short, the degree of...

AI Technical Summary

Problems solved by technology

The characteristics of scattered distribution, various shapes and easy adhesion with other tissues of pulmonary nodules add great difficulty to the detection algorithm.

[0006] The traditional pulmonary nodule detection algorithm uses artificially designed features and classifiers to classify the extracted features. However, pulmonary nodules have various morphological features, and the most important types are hole type and ground glass type. , isolated type, chest wall adhesion type, and blood vessel adhesion type, this differentiation leads to more complex characteristics of pulmonary nodules, and it is difficult to design satisfactory characteristics manually, which makes the detection results still have a large room for improvement

[0007] With the popularity of deep learning algorithms in the field of computer vision, more and more people have begun to try to use convolutional neural networks to classify, detect and segment images, which have achieved amazing results in the field of natural images. Accuracy far exceeds the best hand-designed features

Images