Clinically Applicable Deep Learning to Automate Pulmonary Scanning: Towards Building Intelligent CT Systems

Abstract

Background: Artificial intelligence (AI) has not been experimented with the enhancement of the upstream clinical care, especially imaging acquisition; thus, we hypothesize AI may empower CT with intelligence. We herein combine two state-of-the-art deep learning architectures for object detection and semantic segmentation to build an intelligent system to automate pulmonary scanning and reduce radiation dose. Methods: Facial boundary detection was realized by ingeniously recognizing the adjacent jaw position through training and testing a region proposal network (RPN) on 76,882 human faces after a preinstalled 2-dimensional camera; we then segmented the lung-fields by deploying V-Net on another training set with 314 subjects and calculated the moving distance of scanning couch based on a pre-generated calibration table. A multi-cohort comprising 1,186 patients was used for independent validation and radiation dose quantification under three clinical scenarios. Findings: A U-HAPPY (United imaging Human Automatic Planbox for PulmonarY) scanning CT was designed. The error distance of RPN was 4·46±0·02mm with a success rate of 98·7% in the training set and 2·23±0·10mm with 100% success rate in the testing set. The average Dice's coefficient was 0·99 in the training set and 0·96 in the testing set. A calibration table with 1,344,000 matches was generated to support the linkage between the camera and scanner. This real-time automation makes an accurate planbox to cover exact location and area that needs to scan, which thus reduces amounts of radiation exposures significantly in scenario fully navigated by U-HAPPY as compared to others (all, P<0·001). Interpretation: Our U-HAPPY CT designed for pulmonary imaging acquisition standardization and upstream workflow optimization may complete the last puzzle piece that combines AI and medical imaging. Funding Statement: This work was supported by the National Natural Science Foundation of China (81720108022, 91649116, 81571040, 81973145), the Social Development Project of Science and Technology in Jiangsu Province (BE2016605, BE201707), the National Key R&D Program of China (2017YFC0112801), the Key Medical Talents of Jiangsu Province, the ‘13th Five-Year’ Health Promotion Project of Jiangsu Province (B.Z.2016-2020), the Jiangsu Provincial Key Medical Discipline (Laboratory) (ZDXKA2016020), the Project of the Sixth Peak of Talented People (WSN-138, BZ), the China Postdoctoral Science Foundation (2019M651805), the Fundamental Research Funds for the Central Universities grants of China (Grant No. 2632018FY04), the “Double First-Class” University project (CPU2018GY09). Declaration of Interests: The authors declare no competing interests. Ethics Approval Statement: The study was approved by the institutional review board of the Affiliated Nanjing Drum Tower Hospital of Nanjing University Medical School. Institutional review board (IRB)/ethics committee approvals were obtained. The work was conducted in a manner compliant with the People's Republic of China Health Insurance Portability and Accountability Act (HIPAA) and was adherent to the tenets of the Declaration of Helsinki. Patient consent was waived for access to the training, validation, and independent testing datasets.