Artificial intelligence (AI), neural networks, algorithms, and machine learning (ML) have become powerful tools in the radiologist’s arsenal over the past few years. The evolution of these emergent technologies and their capabilities has allowed for the development of solutions that provide radiologists with reliable insights and guidance that they can use to supplement their professional skill sets. Never a replacement for the expertise of the radiologist, AI has become a safe second pair of hands that can hold some of the weight of the growing imaging workload.
One area where AI has become immensely valuable is in the flagging of pulmonary embolisms (PEs). According to a recent paper released by the American College of Cardiology (ACC) entitled ‘Management of PE’ it is considered to be the third most common cause of cardiovascular death with 60,000-100,000 deaths per year. Treatment of PE depends on its severity and the available resources at the disposal of the healthcare institution.
The Pulmonary Embolism Response Team (PERT) Consortium recently released a consensus document for a determined approach to high and intermediate risk cases using a multi-disciplinary team. The consortium’s goal is to increase awareness of treatment options available to patients with PE and to help minimize worldwide incidence of the condition. In addition to PERT and significant bodies of research, there are detailed pulmonary embolism guidelines to support physicians, radiologists, cardiologists and relevant medical professionals in the identification and treatment of PE. The Management of PE analysis outlined above includes significant detail as does the ACC’s 2019 ESC Guidelines for Acute Pulmonary Embolism published in September 2019.
The use of chest X-rays, V/Q scans, pulmonary angiograms, CT scans, MRIs and duplex ultrasounds to identify PE has meant that a significant portion of the diagnosis load lands on the radiology department screens. In addition to the increase in imaging quantity, there is the need for speed when identifying PE to minimize patient impact and improve outcomes. It’s here that AI is starting to make its mark in supporting radiologists and medical professionals in following pulmonary embolism guidelines.
PE is ultimately life-threatening and can be associated with a myriad of medical conditions from hypertension to obesity to oncology.
At Aidoc, we specialize in the development of AI solutions that can identify acute abnormalities on radiology scans that require urgent intervention. From intracranial hemorrhage to large vessel occlusion or cervical fractures, our solutions are already showing value in a clinical setting. In May 2019, we received FDA clearance for our solution to flag and triage urgent PE cases. One year later, our Incidental PE solution was granted a CE Mark for flagging iPE in contrast-enhanced CT.
At Grupo Fleury, a Brazilian healthcare institution, Dr. Gustavo Meirelles, Medical Manager of Radiology, Strategy and Innovation, was able to expedite a patient for immediate treatment through the use of Aidoc and the flagging of an incidental PE.
“One of our oncologic patient’s lives was saved thanks to Aidoc. The patient underwent a routine chest CT with IV contrast that would only be read in the next couple of days. Aidoc set a priority and there was an incidental PE. We spoke with the clinician and the patient immediately and triaged her to the hospital to be treated – from days to minutes.” he shared.
Few AI solutions focus on incidental findings as they are so low and the specificity of the AI itself has to be extremely high to ensure that the false-positive ratio isn’t too high. Aidoc’s Incidental PE algorithm detects PE exams with lower exam quality – an obvious challenge for any AI. In bypassing these challenges, Aidoc has created the first AI product for triaging incidental findings in medical imaging.
Aidoc’s iPE solution helps to minimize the risk of incidental PEs which can, according to studies, be as high as 2.6%. The solution granted CE Mark and FDA clearance flags incidental PEs in contrast-enhanced CT scans regardless of whether it was elective or emergency and thereby reducing turnaround times for patients with incidental findings to mere minutes. With medical imaging delays due to the COVID-19 pandemic continuing to impact on patient health and care, Aidoc’s iPE flagging, combined with other critical solutions, provides a safety net for healthcare systems under pressure. It’s a support, a helping hand, and a digital eye that never rests for healthcare systems that need to responsibly and efficiently manage urgent and life-threatening conditions while still managing their backlogs.
To fully understand the complexity of PE detection in sparsely annotated 3D CT images, a paper written by Deepta Rajan, David Beymer, Shafiqul Abedin and Ehsan Dehghan goes into the variables that impact PE, the complexities of manual diagnosis and the use of AI to support the detection of PE. The paper entitled ‘Pi-PE: A Pipeline for Pulmonary Embolism Detection Using Sparsely Annotated 3D Images’ offers a solid insight into the complexities of PE and AI and achieving results that can support physician and diagnosis.
To date, the applications of deep learning and algorithms have shown promise and relevance in the support of physicians in adhering to pulmonary embolism guidelines and improved patient care. The world of AI can collaborate with that of healthcare in ways that are both relevant and methodical and, most importantly, capable.