AI in medicine is not a new concept. Over the years, there have been several advancements of the use of AI in medical applications, primarily in developing data-driven support systems for physicians.
Just yesterday, a study published in Nature Medicine reported that a deep learning algorithm was able to boost accurate diagnosis of lung cancer by 5% and reduce false positives by 11%.
A company named Aidoc is among those pushing for the use of AI in radiology. Last week, the company's AI algorithm was granted FDA approval for the identification of pulmonary embolisms in CT scans, close on the heels of the technology receiving a CE mark at the end of February.
Addressing a Medical Need: AI in Radiology
Undetected or late-detected pulmonary embolism, a form of venous thromboembolism (VTE), is one of the most common causes of preventable death in hospitalized patients. According to the CDC, between 60,000-100,000 Americans die of VTE every year—and a quarter of those pulmonary embolism deaths are sudden, without prior symptoms.
Scans Image from Aidoc
Pulmonary embolism doesn’t always present itself in the same manner from case to case, making it difficult for physicians to accurately flag the condition. For this reason, it has been hoped that having an AI-driven workflow triage will help radiologists to pinpoint the condition.
Aidoc's AI solution is used to analyze medical images directly after the patient is scanned. Radiologists are then notified of cases with suspected findings. This serves to assist with prioritization of time-sensitive and potentially life-threatening cases.
The result is that the time from scan to diagnosis for some patients from is cut from hours to less than five minutes, speeding up treatment and improving prognosis.
Efficacy of Aidoc’s Method
Recent research published at the European Congress of Radiology (ECR) in Vienna further shows the accuracy and value that Aidoc's solution can provide.
Within the conference, SS 1404 was entitled “ Advanced lung imaging: MRI and Artificial Intelligence.” The specific topic was formally keyed as: Chest, Artificial Intelligence and Machine Learning, Imaging Methods.
The specific subsection of concern was “AI-powered detection of pulmonary embolism in CT pulmonary angiograms: a validation study of the diagnostic performance of prototype algorithms.”
A clinical case study published on Aidoc's website entitled "AI-Powered Detection of Pulmonary Embolism in CT Pulmonary Angiograms: A Validation Study of the Diagnostic Performance of Prototype Algorithms" reports impressive findings: "The best performing algorithm [in the test] was a fully convolutional neural network with a backbone based on the Resnet architecture. It achieved a sensitivity of 93% and a specificity of 95%. This corresponds to a positive predictive value of 77%."
Always-On Artificial Intelligence
As described by Elad Walach, Aidoc co-founder and CEO, Aidoc’s AI software can be described as “always-on AI”. He goes on to say that, contrary to the prevailing “on-demand AI” where a doctor has to request the intervention of the AI solution, “Always-on AI” works in the background to keep radiologists focused on making diagnoses.
In this manner, when an abnormality is detected, physicians will be notified, directly in their worklist. Without naming names, Walach notes that a Chairman of Radiology told him that “you gave me the peace of mind that there were no patients with a brain bleed waiting for their scans to be read.”
Validating AI Solutions in Radiology
The use of AI in radiology has grown from an academic curiosity to a part of the clinical process. Recently the Data Science Institute (DSI) of the American College of Radiology (ACR) announced its AI validation process.
As part of that process, ACR DSI has begun its process of building out its Assess-AI process and registry. Assess-AI will monitor of AI algorithm performance in clinical practice by capturing real-world data during clinical use and installing it in a clinical data registry. Aidoc and Nuance Healthcare will be part of a process to “test and refine the processes for data gathering”, as ACR DSI puts it, at a project taking place at University of Rochester Medicine.
To quote ACR DSI, “By providing developers with longitudinal algorithm performance data, we provide a pathway for meeting any FDA post-market surveillance requirements and specific information developers can use for algorithm improvement.”
Aidoc’s Other Medical AI Efforts
The pulmonary embolism-related AI workflow technology is just the most recent news from Aidoc. As of this writing, the company has eight more technologies in active clinical trials.
Aidoc’s technology has been approved in various cases by the FDA, though limited by United States and EU law to "investigational use". Image from Aidoc
In August 2018, Aidoc also received FDA approval for an AI solution that works with radiologists to flag acute intracranial hemorrhage cases in head CTs.
While AI is not going to replace doctors any time soon, it is increasingly providing advanced tools to enable healthcare professionals to make better healthcare decisions. Perhaps the question will then become not whether you'd trust an AI to provide you healthcare—but rather would you trust a doctor to treat you without AI?