Deep Learning Based Common Thoracic Disease Classification And Localization In Chest X-ray

Thoracic disorders are conditions of the heart,lungs,mediastinum,esophagus,chest wall,chronic obstructive pulmonary disease,cystic fibrosis diaphragm and great vessels.Thoracic diseases are one of the most common diseases in the world,and is considered the leading cancer in causing death worldwide.Early detection and diagnosis of thoracic disease can greatly improve patient treatment during the early stage and hence potentially increase the survival rate and patients’ quality of life.The early stage detection of thoracic disease can potentially reduce cancer mortality.The chest X-ray is one of the most reachable radiological examinations for screening and diagnosis of thoracic diseases.A number of X-ray imaging studies lead by radiological reports are assemble and stored in many modern hospitals.The one thing is how this type of hospital-size knowledge database containing invaluable images information(i.e.untied labeled)can be used to facilitate the data deep learning paradigms in building large-scale high precision computer-aided diagnosis systems.Detecting thoracic diseases on chest X-rays is a difficult task,due to the highly varied appearance of contusion areas on X-rays from patients of different thoracic disease.There is also the problem of shortage of accurate pixel-level annotations by radiologists for model training.Existing machine learning procedure are unable to deal with the challenge that thoracic diseases usually happen in localized disease-specific areas.We manifest a weakly supervised deep learning model for classifying diseases and identifying abnormalities based on medical imaging data.The process of learning from medical imaging data with various thoracic lesions,the model was trained on imaging data with image level labels to classify diseases,and is able to identify abnormal image regions simultaneously.The model consist of a pooling structure and a Dense Net front-end,which can recognize possible disease features for classification and localization tasks.Our model has been validated on Chest X-ray14 dataset.The structure can produce accurate disease classification and localization,which can potentially support clinical decisions.A weakly supervised deep learning framework also equipped with multi-map transfer and max-min pooling for classifying thoracic diseases as well as localizing suspicious lesion regions.