Thanks to SMGR Radiologist, Dr. Jesse Rael, for a thought-provoking Grand Rounds titled AI in Radiology this week. Super interesting.
The recording is available HERE for your viewing pleasure.
Here are my thoughts on Dr. Rael's presentation: facial recognition at airport (cnn.com)
Artificial intelligence is the the theory and development of computer systems able to perform tasks that normally require human intelligence, such as visual perception, speech recognition, decision-making, and translation between languages.
AI is present in many aspects of modern life, including facial ID, social media platforms, google search, digital voice assistants (siri and alexa), smart home devices, amazon recommendations, and more.
The topic of AI feels very far from where we are in the world of safety net primary care-- where literally just getting a patient a follow up appointment with a PCP feels like moving a mountain. But I am not sure my sense is actually accurate. In fact, Dr. Manny Mendoza gave a Grand Rounds just a few months ago on AI in Primary Care. The link to that GR summary and recording is here. And the truth is, AI is already making changes in primary care: decision-making tools, benign vs. malignant lesions in dermatology, chronic disease management and more.
With regards to radiology, AI means a computer that is trained to interpret images to either rule in/out a radiographic diagnosis. This could include a wide range of diagnoses, from fractures to pulmonary emboli to cancer, and beyond.
Proponents of AI in radiology argue that radiologists are expensive, there is increasing expectations of productivity, that imaging is getting more complex, numbers of images per study are increasing, and that AI could streamline and improve that work.
Dr. Rael doesn't think that computers are going to replace radiologists, BUT, he believes, radiologists who don't adopt the evolving technology will likely be replaced. In his words, AI could be considered equivalent to autopilot in a jet engine cabin-- the pilot is absolutely needed, particularly for complex situations, and the autopilot function is there to help him do his job better.
Dr. Rael's hope is that AI will help radiologists to become more productive, enhance protocols, be available where radiologists are not, and more. And, as he points out, radiologists do much more than interpret images-- they are involved in clinician-to-clinician and clinician-to-patient communication, QA, education, policy making, hands on scanning, biopsies, etc.
Dr. Rael spent some time helping us to understand the concept of "deep learning" currently being used in image interpretation. He showed us how a computer can "see" images (e.g. a kidney or a brain lesion) on a screen.
There are many different current AI projects around the world: identification of pulmonary edema on CXR, white matter abnormalities in very preterm infants, PE recognition, benign vs. malignant breast lesions, abnormalities in knee MRI, and automatic scan range delimitation in Chest CT. The possibilities are endless! Dr. Rael himself is involved in a neuroradiologic project looking at image analysis for different brain lesions-- what does a glioblastoma look like? How can a computer recognize it?
And what about global health implications for providers working in places where there are not only no radiologists, but no machines to image (e.g. butterfly ultrasound with images being interpreted by offsite AI)?
What do you think about AI in radiology and AI in general in medicine? Deeply interesting, kind of scary, and definitely cool.
My advice for this week: if you have a question about a radiographic finding, don't forget to call your friendly local radiologist-- I always feel like I understand more about a complex impression, after I have spoken to a real live human radiologist. Will that always be so?
No comments:
Post a Comment