Artificial intelligence is the next big thing in radiology, as well as in many other branches of medicine. Whether it will assist us or replace us in the future remains a matter of intense debate, but how many radiologists really understand basics of how AI actually works and how accurate can it get? Understand the basics of AI as I review a landmark article on Artificial Intelligence in imaging acute CT brain recently published in Lancet.
You can read the entire Lancet article here.
– Akshay Baheti, Tata Memorial Hospital
PS : If the embedded video is not opening, you can view the video on YouTube by clicking here.
4 thoughts on “Artificial Intelligence in Radiology: The how and the why of it”
Pingback: Artificial Intelligence in Radiology: The how and the why of it. | SOMEONE SOMEWHERE
Superb post as always Dr. Akshay!
How long do you think AI will take to get integrated into the current radiology workflow, in India and abroad?
Thanks Amar! It’s already happening in many institutes. It’s integration will probably first happen at the level of work flow optimization. For image interpretation, ‘mundane’ jobs like reading chest xrays and emergency situations like ct brain for acute trauma or stroke will probably get AI options soon. For tumors and complex stuff, there will be an increasing AI component which predicts things based on texture analysis etc, but this technology will take years to mature, and won’t happen on a broad scale level so soon. The softwares which get integrated into Siemens or GE or Phillips directly (such that they are offered along with the CT or MRI purchase itself) will probably be the fastest to integrate, compared to neutral third party vendors which will need a separate level of integration into the hospital pacs etc. But things will change pretty much every year now onwards! How are things in Canada?
To add further, a study just published in KJR just proves the point I made in the video about the quality of the AI studies published. Only 6% of the AI studies reviewed by the authors had an external validation dataset as per the study. You can read more details at https://doi.org/10.3348/kjr.2019.0025