Waiting for medical report results is often a long and agonizing process for patients. Now, Wan-Yuo Guo, Chief of the Department of Radiology at Taipei Veterans General Hospital, has introduced an AI-assisted outpatient service that drastically shortens the diagnosis time for brain tumors. Patients can now learn their condition on the very day of their examination, overturning the prediction that radiologists are destined to be replaced by AI.
 

Just last year, Kai-Fu Lee, Chairman of Sinovation Ventures, named radiologists as one of the top ten professions most likely to be replaced by AI in his book AI Superpowers. However, Guo seems undeterred. He established the "AI Neuro-imaging Assisted Outpatient Clinic" specifically to challenge this prophecy.

 

Drastic Reduction in Diagnosis Time

 

Dr Guo explained the workflow changes introduced by AI. Currently, confirming whether a patient has lung cancer involves a three-step process arranged by a chest specialist: a chest Computed Tomography (CT) scan, a brain Magnetic Resonance Imaging (MRI) scan, and a nuclear medicine bone scan.

 

Because the lungs have a unique structure, small cancer cells can readily travel through the bloodstream to the brain. Therefore, standard lung cancer staging requires a brain CT or MRI. However, scheduling an MRI usually involves a wait of at least one month. After the scan, it takes an additional 5 to 7 days for the report to be generated, followed by a second outpatient appointment where a specialist determines the subsequent treatment. The entire process takes approximately 1.5 months to complete.

Conventionally, a radiologist must sift through hundreds of MRI images, manually circling lesions one by one and calculating tumor volume to determine if surgery is necessary. It typically takes a radiologist with four years of residency training, plus two years of subspecialist experience, about 20 minutes to perform such an interpretation.

 

In contrast, the "DeepMets" clinical AI automated detection system for brain metastases—jointly trained by Dr Guo's team and the Taiwan AI Labs—allows physicians to view preliminary interpretation results on a computer in under 30 seconds.

 

AI Accuracy Comparable to a Resident Physician

 

How does the AI's interpretation compare to that of a senior physician? Dr Guo tested the system on the spot using the MRI results of an anonymous patient. Using the same batch of MRI images, Dr Guo identified two brain metastases, while DeepMets marked at least five "regions of interest" suspected to be tumours.

"It is not 100% accurate right now, but I would rather it have higher sensitivity," said Dr Guo.

Currently, DeepMets is still in the learning phase, with an accuracy rate of approximately 85%, which is comparable to that of a medical resident. "Determining whether the diagnosis provided by the AI algorithm is correct in the early stages still relies on experienced senior physicians to act as gatekeepers," Dr Guo noted.

He believes that talent capable of utilizing AI will remain competitive. "In the future, it will likely be doctors who know how to use AI replacing those who don't."
 

Read the original article: https://www.cw.com.tw/article/5096581?template=transformers

Back to the top