A method for suppressing false positive samples of pulmonary nodules based on 3dcnn

Abstract

The invention discloses a pulmonary nodule false positive sample inhibition method based on 3D CNN (convolutional neural networks). The pulmonary nodule false positive sample inhibition method comprises steps that 1) 3D reconstruction of a lung CT is carried out by adopting interpolations; 2) according to a to-be-inhibited sample coordinate, a cube of a fixed size is cut from reconstruction data, and is normalized, and then positive samples are expanded, and then negative samples and positive samples are used as the training data of the 3D CNN together; 3) a training sample is used to train a 3D CNN model; 4) during the model training, weighting correction of a standard loss function is carried out, and different weights are provided for the negative samples and the positive samples. A training network is cycled and iterated, and finally, the 3D CNN model is acquired. The method based on the 3D CNN training model is advantageous in that on one hand, the three-dimensional characteristics of the lung CT data are effectively used, and sample information is reflected to the greatest extent; and on the other hand, the weighting of the loss function is carried out during the model training, a problem of true and false sample imbalance is solved, and therefore the model having good pulmonary nodule identification effect is acquired after the training.

Description

technical field [0001] The invention belongs to the field of intelligent diagnosis of medical images, and in particular relates to a method for suppressing false positive samples of pulmonary nodules based on 3DCNN. Background technique [0002] The detection of pulmonary nodules is critical to the processing of lung CT images, which is a major manifestation of lung cancer in its early stages. Effective early detection and screening of pulmonary nodules can significantly improve the five-year survival rate of lung cancer patients, so it has very important research value and significance. [0003] Although the emergence and development of CT imaging technology and various new diagnostic and detection methods, as well as the emergence of various new CT technologies, the diagnosis of lung cancer has become relatively easier than before, but it is still difficult to detect cancer at an early stage. Moreover, new CT technologies, such as multi-slice CT to produce a huge number o...

AI Technical Summary

Problems solved by technology

However, the distinction between true and false positive samples of pulmonary nodules is not obvious, and the task of manual selection and design of features that can be distinguished is complex, often requiring researchers with rich professional knowledge for several years of research to select a suitable sample. The characteristics of the task requirements, establish a classifier; once the task changes, the selected and designed features will become invalid, and new features need to be selected and designed according to the characteristics of the new task

Such research consumes a lot of manpower and material resources and cannot achieve satisfactory results.

Images