Method of automatically detecting and sketching pulmonary nodule positions based on convolution classifier

Abstract

The present invention relates to a method for automatically detecting and delineating the location of pulmonary nodules based on a convolutional classifier, including: (1) delineating and storing the positions of pulmonary nodules and suspected pulmonary nodules, marking and storing pulmonary nodules (2) check whether the stored information in step (1) is correct; (3) the position of the pulmonary nodule and the position of the suspected pulmonary nodule outlined in step (1) are the ROI area, Perform translation, scaling, rotation, compound rotation, and a combination of two or more of translation, scaling, rotation, and compound rotation, and use all the obtained operation results as a sample set; (4) input some samples in the sample set into the convolution The neural network classifier outputs the correct ROI area where the pulmonary nodule is located and the correct ROI area where the suspected pulmonary nodule is located. With the data set obtained by this method, the detection accuracy rate is increased by more than 5%, and the false positive is reduced by more than 0.4%.

Description

technical field [0001] The invention relates to a method for automatically detecting and delineating the location of pulmonary nodules based on a convolutional classifier, and belongs to the fields of computer vision-based pulmonary nodule detection, medical image retrieval and the like. Background technique [0002] Lung cancer is considered to be the number one killer threatening human health because of its extremely high morbidity and mortality. The early manifestations of lung cancer on medical imaging are usually solitary pulmonary nodules. In order to objectively provide doctors with suspicious pulmonary nodule markers, a lung CAD system was introduced to assist doctors in diagnosis. After the computer diagnosis system automatically diagnoses and analyzes lung CT images, it prompts doctors for suspicious pulmonary nodules in CT images, thereby Overcome some subjective factors of doctors in diagnosis and improve the detection rate of lung cancer. [0003] In the tradi...

AI Technical Summary

Problems solved by technology

However, the number of training samples in this patent is too small to achieve the desired effect, which directly affects the accuracy of the test; and the patent does not mention how to extract lung nodules from CT images;

[0005] In the existing technology, training a deep convolutional network requires a large number of training samples. If it is obtained through software technology, the correctness of the obtained image is questionable. The correctness of the training samples directly affects the accuracy of the classification results.

If all samples are obtained manually, the accuracy is guaranteed, but this is a time-consuming and labor-intensive task

Images