AI in CXR reporting
23 Jan 2020
Baseline audit of CXR reporting AI in a DGH setting
ChestEye is an AI diagnostic which is able to produce a report on chest radiographs received as DICOM. By default it makes a call on 75 different findings. A retrospective baseline audit was made of ChestEye’s capability to detect normal films using the original verified report as a criterion. Based on these findings, recommendations are made for its subsequent deployment for a trial period alongside a newly trained reporting radiographer within the CQC regulatory sandbox for AI in radiology.
Educational aims:
• To learn how does CXR AI perform in the detection of normal films in a DGH setting
• To find out what the potential implications of this performance for improving report turnaround time for A&E CXR
• To understand how can regulatory bodies encourage the appropriate safe early adoption of diagnostic AI in the UK
References:
• Yasaka K, Abe O. Deep learning and artificial intelligence in radiology: Current applications and future directions.PLoS Med. 2018 Nov 30;15(11):e1002707. doi: 10.1371/journal.pmed.1002707. eCollection 2018 Nov.
• Shiraishi J, Katsuragawa S, Ikezoe J, Matsumoto T, Kobayashi T, Komatsu K, Matsui M, Fujita H, Kodera Y, Doi K.Development of a digital image database for chest radiographs with and without a lung nodule: receiver operating characteristic analysis of radiologists’ detection of pulmonary nodules.AJR Am J Roentgenol. 2000 Jan;174(1):71-4.
• Care Quality Commission Sandbox round 2: Machine Learning (AI) and its use in Radiology,Pathology and other similar diagnostics.
• https://www.cqc.org.uk/sites/default/files/20190911_Round2-AI_radiology_diagnostics_screening_services.pdf
Further reading:
• https://www.oxipit.ai/
ChestEye is an AI diagnostic which is able to produce a report on chest radiographs received as DICOM. By default it makes a call on 75 different findings. A retrospective baseline audit was made of ChestEye’s capability to detect normal films using the original verified report as a criterion. Based on these findings, recommendations are made for its subsequent deployment for a trial period alongside a newly trained reporting radiographer within the CQC regulatory sandbox for AI in radiology.
Educational aims:
• To learn how does CXR AI perform in the detection of normal films in a DGH setting
• To find out what the potential implications of this performance for improving report turnaround time for A&E CXR
• To understand how can regulatory bodies encourage the appropriate safe early adoption of diagnostic AI in the UK
References:
• Yasaka K, Abe O. Deep learning and artificial intelligence in radiology: Current applications and future directions.PLoS Med. 2018 Nov 30;15(11):e1002707. doi: 10.1371/journal.pmed.1002707. eCollection 2018 Nov.
• Shiraishi J, Katsuragawa S, Ikezoe J, Matsumoto T, Kobayashi T, Komatsu K, Matsui M, Fujita H, Kodera Y, Doi K.Development of a digital image database for chest radiographs with and without a lung nodule: receiver operating characteristic analysis of radiologists’ detection of pulmonary nodules.AJR Am J Roentgenol. 2000 Jan;174(1):71-4.
• Care Quality Commission Sandbox round 2: Machine Learning (AI) and its use in Radiology,Pathology and other similar diagnostics.
• https://www.cqc.org.uk/sites/default/files/20190911_Round2-AI_radiology_diagnostics_screening_services.pdf
Further reading:
• https://www.oxipit.ai/
Duration:13 mins
Speaker info
Adrian Tang
Dr Adrian Tang is a Consultant Radiologist at East Cheshire NHS Trust After medical training at The Queen’s College, Oxford, Dr Adrian Tang completed his MRCP. His radiology CCST was started in Leeds and completed on a three year Translational research Fellowship at The Royal Marsden Hospital/ICR under Professor Janet Husband. A key role was providing novel and routine imaging support to the fledgling Drug Discovery unit with Professors Kay, Judson and de Bono. MR and CT protocol design, percutaneous biopsy Cancer MDT support and physics liaison in a translational environment were key skills and philosophies acquired. He maintains an enthusiasm and affinity for translational collaboration between frontline science, industry and the clinical coalface. He believes that this brings imagination and opportunity to education and training for medical students through radiography colleagues to radiologists at all levels. The current collaboration with Oxipit AI has been accepted on CQC Regulatory Sandbox for artificial intelligence in radiology and adds experience with national regulatory agencies.