Subsolid nodule quantitative analysis method and system based on image depth characteristics

Abstract

The invention discloses a subsolid nodule quantitative analysis method and system based on image depth characteristics. The method comprises following steps of S1, acquisition and preprocessing of a lung CT image; S2, candidate lung nodule ROI segmentation and extraction; S3, using a convolution neural network to extract subsolid nodule image depth characteristics; S4, constructing an ELM prediction model based on group sparse constraints, introducing an L1 norm, characteristic selection stability is improved and overfitting is prevented. By adding group sparse constraints, model robustness and generalization ability are improved. According to the invention, by use of subsolid nodule CT image characteristics, pathology (before infiltration, micro-infiltration and infiltration) is predicted. By extracting the image depth characteristics of the subsolid nodule, quantitative analysis of pathology aggressivity can be finished.

Description

technical field [0001] The invention relates to the field of image analysis, in particular to a method and system for quantitative analysis of subsolid pulmonary nodules based on image depth features. Background technique [0002] Lung cancer is the leading cause of cancer death worldwide. Although the 5-year survival rate of lung cancer is low, early diagnosis and treatment of lung cancer can significantly improve the 5-year survival rate after surgery. Studies have found that subsolid pulmonary nodules are closely related to early lung cancer. With the increasing number of routine CT examinations and low-dose CT screening of the lungs, the detection rate of subsolid pulmonary nodules is increasing. The traditional diagnosis methods of subsolid pulmonary nodules mainly rely on the experience of radiologists, and there are problems in quantitative analysis. How to use cutting-edge image processing technology to quantitatively extract the image features of subsolid pulmonar...

AI Technical Summary

Problems solved by technology

However, subsolid pulmonary nodules are difficult to ensure the stability of the extracted features due to their blurred edges, irregular shapes, and large random changes in solid components, which brings difficulties to quantitative analysis.

[0004] Traditional feature extraction of pulmonary nodules often defines a large number of features, and selects the features most relevant to pathological erosion for erosion prediction. The traditional feature selection uses the LASSO method. The disadvantage is that 1. The space shows the characteristic of structured "cluster" prior. LASSO only considers the sparsity and ignores this structured prior, resulting in poor stability of the selected features; The amount of calculation is large; 3. The LASSO method is based on a linear model and does not consider nonlinear factors

Images