Evaluation of Pneumoconiosis High Risk Early Warning Models

Overview

Precaution of pneumoconiosis is more important than treatment. However, the current process can't early warn the high-risk dust exposed workers until they are diagnosed with pneumoconiosis. With the feature of efficiency, impersonality and quantification, artificial intelligence is just appropriate for solving this problems. Therefore, we are aiming at adapting deep learning to develop models of pneumoconiosis intelligent detection, grade diagnosis and high risk early warning. The annotated images will be used for convolutional neural networks (CNNs) algorithm training, aiming at pneumoconiosis screening and grade diagnosis. Moreover, risk score calculated by density heat map will be used for early warning of dust-exposed workers. Then follow up of cohort will be implied to verify the validity of the risk score. By this way, the high-risk dust-exposed workers will get early intervention and better prognosis, which can obviously reduce medical burden.

Pneumoconiosis, the predominant occupational disease in China and all over the world. Chest radiography is the most accessible and affordable radiological test available for the physical examination of dust-exposed workers and mass screening for pneumoconiosis. But the diagnosis process has some disadvantages, such as strong subjectivity, inefficiency, and disability of judgement of borderline lesion, etc. Besides, precaution of pneumoconiosis is more important than treatment. However, the current process can't early warn the high-risk dust exposed workers until they are diagnosed with pneumoconiosis. With the feature of efficiency, impersonality and quantification, artificial intelligence is just appropriate for solving the aforesaid problems. Up to now, there has been rare research about adapting deep learning for pneumoconiosis grade diagnosis and high risk early warning. In our previous studies, we set up a chest radiograph database, which contains more than 100,000 digital pneumoconiosis radiography images. The result of detection-system evaluation demonstrated that the accuracy in the identification of pneumoconiosis could reach 90%, with an AUC(Area Under The Curve) of 0.965 and a sensitivity of 99%. More works need to be continued. Therefore, we are aiming at adapting deep learning to develop models of pneumoconiosis intelligent detection, grade diagnosis and high risk early warning. The annotated images will be used for convolutional neural networks (CNNs) algorithm training, aiming at pneumoconiosis screening and grade diagnosis. Moreover, risk score calculated by density heat map will be used for early warning of dust-exposed workers. Then follow up of cohort will be implied to verify the validity of the risk score. By this way, the high-risk dust-exposed workers will get early intervention and better prognosis, which can obviously reduce medical burden.