AI in radiology: A view of the immediate future

AI in radiology: A view of the immediate future

Artificial intelligence is rapidly making its way into virtually all spheres of daily life. Itโ€™s certainly no different for radiologists, who are currently faced with unprecedented advancements in the field of machine learning and AI in radiology.

So how long will it take until human radiologists are redundant? Will we ever reach a point where machines can handle diagnostic imaging without making mistakes? And how much can machines do in clinical practice today?

SCP Managing Partner Dr Jean de Villiers was recently interviewed on CapeTalkโ€™s Afternoon Drive and RSGโ€™s Monitor to summarise, briefly and in laymanโ€™s terms, what South African radiologists are navigating at present.

.

.

Key takeaways from the conversations include:

1. AI has been part of radiological technology for a while

Radiologists have successfully used forms of AI for many years, notably to improve workflow in their practices. โ€œIn the simplest form, itโ€™s been used in compiling and allocating work lists to radiologists in practice around the world, as well as to synchronise prior examinations if we are reading follow-up oncology or cancer studiesโ€ฆ so we can compare them accurately.โ€

Against the background of a global radiologist shortage and the pressure of having to report increasing numbers of images, something as simple as a โ€œsmartโ€ work list that delivers relevant studies to each subspecialist, has an invaluable time-saving effect.  

2. The diagnostic use of AI in radiology isnโ€™t brand new but improving rapidly

โ€œThe diagnostic use of AI has been around for years already but it only recently became more robust with the advent of more powerful computers,โ€ Dr de Villiers says. โ€œAlso, with the acknowledgment and passing of the FDA hurdles in the use of it.โ€

Note: Of the 521 AI/machine learning-enabled medical devices approved by the FDA since 1995, 415 have been approved in the last five years. Three quarters of the devices are used in radiology.

โ€œAt the moment we can use it in a practical form to detect brain bleeds, for instance. We can use it to detect pneumothorax which is when air leaks from the lungs, we can use it to identify blood clots in the pulmonary arteries and also to detect early breast cancers.

โ€œI think the next major benefit of AI will be in the triage of patients. If we have ten or twelve CT patients, and one of them has a life-threatening brain bleed which needs to be reported urgently, AI will have the ability to detect it on our behalf almost before it appears on our reporting list.โ€

3. Even considering the rapid progress, AI still only provides part of the answer

โ€œI remember back in 2006โ€ฆ we installed a multidetector CT and with it came the ability to detect lung nodules. At the time it took ages to do and it wasnโ€™t all that accurate.

โ€œ[Today] there certainly are AI modules that assist us in detecting nodules in the lungs. But bear in mind that most of the AI tools we have are specifically aimed at an answer; not the whole answer but part of the answer.

โ€œIf you came into the practice and you had a CT of your lungs, those images might be run through an algorithm which would be for detecting nodules. That algorithm would not pick up consolidation which might be an indication of infection. It wonโ€™t necessarily pick up interstitial scarring or some sort of parenchymal distortion in your lungs because itโ€™s aimed only at detecting nodules.

.

.

โ€œThere might be another algorithm which would detect clots within the arteries in your lungs (called pulmonary embolism), but again thatโ€™s part of the equation and not the whole equation. At this point the whole equation is still very squarely in human hands.

โ€œThe algorithms are diagnostic assistants, as it were, to try and help us first of all to be more accurate and also to try and alleviate a bit of the volume and stress that we as radiologists often have as a result of work volumes.โ€

4. Humans will remain part of the diagnostic process for the foreseeable future

Dr de Villiers believes that one can never know what the future holds but at this stage, we are very far from a place where humans will be replaced by AI in radiology.

โ€œIโ€™ve been in practice for 20 years plus and on a daily basis I see things that Iโ€™ve never seen before; abnormalities that need investigation and at the moment the algorithms and machine intelligence that weโ€™re using is dependent largely on whatโ€™s fed in.โ€

As diseases evolve, new information will need to be added to the algorithms for them to know which abnormalities to detect.

โ€œGoing forward, Iโ€™m very much cautious but excited about what weโ€™re dealing with. For a long time, there will be human intervention required. Firstly, because at the moment AI brings only part of the equation and, secondly, inevitably the remit or the job of the radiologist is to interpret what the AI module might be producing.

โ€œProbably, going forward, together we’re better than either AI or humans on their own.โ€

.

*SCP Radiology provides medical imaging services at 19 branches in the Western Cape, including MRI, CT, mammography, X-rays and others. For more information on our diagnostic imaging services, click here.