Pneumoconiosis identification and classification method based on deep convolutional neural network

Abstract

The invention discloses a pneumoconiosis identification and grading determination method based on a deep convolutional neural network, which includes: obtaining and collecting digital chest radiographs and information of pneumoconiosis patients who have been retrieved and screened; obtaining and collecting age, gender and The above-mentioned pneumoconiosis patients match the digital chest radiographs of normal people; prepare corresponding data samples for the training of deep convolutional neural network; obtain the pneumoconiosis judgment and classification model through training; input the digital chest radiographs to be judged and graded to the In the pneumoconiosis judgment and grading model, output the probability of judging whether pneumoconiosis is present; generate and output the class activation heat map; compare and score the small shadow shape obtained according to the statistics with the corresponding standard digital chest radiograph, and obtain the relative density value; according to The relative density value outputs the determination result of pneumoconiosis stage. By adopting the present invention, the judgment and grading of pneumoconiosis are not only intuitive but also give reasonable judgment reasons, the effect is significantly improved and it is safer and more reliable.

Description

technical field [0001] The invention relates to the technical field of information analysis, in particular to a pneumoconiosis identification and grading determination method based on a deep convolutional neural network. Background technique [0002] Pneumoconiosis is a collective term for a group of occupational lung diseases mainly caused by diffuse fibrosis of lung tissue caused by long-term inhalation of different pathogenic productive dust and retention in the lungs during practicing activities. There are many types of powders that cause pneumoconiosis. According to the "Classification and Catalog of Occupational Diseases", it mainly includes silicosis, coal workers' pneumoconiosis, graphite pneumoconiosis, carbon black pneumoconiosis, asbestosis, talc pneumoconiosis, cement pneumoconiosis, mica pneumoconiosis, potter's pneumoconiosis, aluminum pneumoconiosis, welder's pneumoconiosis There are twelve kinds of pneumoconiosis and castor's pneumoconiosis. The number of pa...

AI Technical Summary

Problems solved by technology

There are the following problems: ① Strong subjectivity: Affected by various factors such as the experience of the film reader and the conditions of the film reading, the film reading cannot be standardized, and the conclusions drawn by different institutions, different doctors or even at different times may fluctuate, and young doctors Often inexperienced, the accuracy of film reading is not high; ②difficult to quantify: It is impossible to accurately judge critical lesions (especially pneumoconiosis and first-stage pneumoconiosis), and the diagnosis of pneumoconiosis and first-stage pneumoconiosis is very important for the identification of work-related injuries in patients ;③Loss of original information: Diagnosis mainly depends on developing film, but there are information loss in the process of image transmission and processing, which is greatly affected by different equipment conditions;④High cost: low efficiency, unable to process in batches, requires the participation of multiple doctors Diagnosis, high labor cost; need to print chest X-ray, high cost of consumables and storage

In 2014, Zhu et al. also applied the above technique for the diagnosis of pneumoconiosis, but the amount of data was small (85 people in the normal group, 45 people in the pneumoconiosis group)

In 2015, Luo Haifeng improved the technology of BP artificial neural network for the interpretation of pneumoconiosis chest X-ray shadow density, which was used for pneumoconiosis staging, but the average correct rate of classification was only 68.3%

The limitations of the above research are the small amount of data and the low accuracy of the conclusions. In addition, using the ANN method still requires feature engineering, and the model is a shallow neural network, which cannot perform effective representation learning to achieve the desired accuracy.

[0009] In 2017, Deng Kui applied a deep convolutional network (GoogleNet) to the diagnosis of pneumoconiosis chest X-ray, and the effectiveness was 91.6%. However, the traditional image technology was still used in the graded diagnosis, and the process was cumbersome. Significant gaps in tasks compared to state-of-the-art models

Most of these studies focus on the identification of pneumoconiosis, and the amount of data is small, the algorithm used is relatively old, and the diagnostic efficiency and accuracy are low

Images