Radiology Residents of the Future: Dr. Atin Saha

This series highlights the insights and learnings from medical residents engaged in the use of artificial intelligence for their medical practice

Over the past few years, artificial intelligence (AI) has slipped into mainstream conversation and application, providing insights that support the radiology profession in managing complex and overpowering workloads. The technology is being constantly refined and adapted to provide increasingly detailed and accurate analyses and to meet the exacting standards of the medical profession. For Dr. Atin Saha, Junior Chief Resident at Yale, AI has become a reliable tool that supports not just clinical diagnosis in busy periods, but a teaching aid that gives residents insights into their performance and success rates.

AIDOC: Can you tell us a bit more about yourself?

Dr. Atin Saha: I did my Masters in Biomedical Engineering. It was at this point that I was introduced to machine learning algorithms in the application to identify neural maps when solving puzzles. So, my first foray into AI was from an engineering standpoint, then I went to medical school at Brown and become very interested in radiology. This was informed by my background in engineering and early research work in the field.

After I completed my studies at Brown, I came to Yale for my radiology residency where I was introduced to Aidoc on the clinical side in my first year. Then, in my second year, it was fully integrated into the service, particularly for head and cervical spine imaging. This ignited my interest in how AI could be used to help resident education and I started working with my colleagues to see if we could use these tools already utilized in the clinical workflow to help residents learn better. We got involved with the Aidoc team and collaborated on building a platform that allowed us to test these theories. In the pilot study, we found that AI helps resident physicians improve accuracy and demonstrated that use of AI as a learning platform has immense value.

AIDOC: How does this work with the residents and the profession? What value does it add?

Dr. Atin Saha: The effects are twofold. The first is that the more you read and study, the better you are at improving your accuracy as you’re assessing a larger volume of cases. The second is that by using Aidoc, you are picking up feedback about your performance in real-time. Residents learn as they go through the cases – they can see what they think, decide if they agree or disagree and then Aidoc gives them immediate feedback. This means that they can learn a lot faster.

AIDOC: How does AI support you as a radiologist?

Dr. Atin Saha: I believe that the training that a radiologist undergoes is the essential foundation for any clinical diagnosis. AI offers an additional layer of support that helps with this diagnosis and the workload. The way I see it, Aidoc flags things that have been found, like an intracranial hemorrhage and then the radiologist decides if they agree with the system’s assessment. Perhaps the best way to think of it is that if Aidoc wasn’t there, you’d only have one pair of eyes reading the images, but with Aidoc it is two pairs of eyes. It makes it easier to go through flagged images to see if I have a lead early on in the process or the day.

AIDOC: Does Aidoc offer you peace of mind when working night shift or helping junior residents?

Dr. Atin Saha: I think it’s a good thing. You can use it to highlight items such as a bleed or a fracture, really early on. I had a conversation about a month ago about how this is very helpful as it puts flags onto the reading list that I can immediately look at and assess. It adds an extra sense of confidence to the workload, knowing that someone else has already looked for a bleed or a fracture or emergency issue and flagged it up for checking. Ultimately, the more people who look at a case, the better it is for the patient and the case.

AIDOC: Was AI in this use case a surprise to you?

Dr. Atin Saha: I honestly didn’t expect to use AI in at this level by the time of my residency, maybe in the middle of my career, but definitely not this early on. I think this is playing no small part in it becoming such a hot topic. A lot of people assume that it’s there to make the diagnosis and replace the radiologist rather than to assist in making the diagnosis. That’s the big difference. For me Aidoc is that extra pair of eyes that helps me make sure I don’t miss anything rather than the tool I use to make a diagnosis. At the end of the day, I am the one who makes the interpretation.

AIDOC: What does your day-to-day look like now that you use Aidoc?

Dr. Atin Saha: When I start my ER shift, I log into my computer and as the list is compiling, I take a look at the list from Aidoc on the left-hand side of my screen. I check to see if there’s anything flagged by Aidoc, then I open the study to see if it is a positive match or if I disagree. This then helps me to prioritize what’s on the list and integrates neatly into the ER workflow. Currently, we have Aidoc in ER, neuro, and chest.

AIDOC: Can you tell us more about the AI residents project?

Dr. Atin Saha: The project was designed to see if Aidoc would be able to help residents quantify sensitivity and accuracy in terms of their ability. To give you an example – unless you physically write down the number of cases you assessed and how well, there are no numbers that quantify your success rate. Yet, through Aidoc and this study, we can now provide residents and practicing radiologists with tangible information about their performance. For example, the system can tell them that they are 95% accurate in assessing intracranial hemorrhage on non-contrast head CTs. This is invaluable. You can spend four years finishing your residency and not have a good quantitative insight into how you work. This is a significant outcome for this project.

The project includes a combination of non-contrast CT-head and cervical spine studies and each resident was asked to classify whether these were positive or negative. They were then presented with close to 200 cases of which half were analyzed by Aidoc and half not. Sensitivities increased with the use of Aidoc for all levels of residency training. It provides radiology residents with real-time feedback and we’ve seen huge percentage increases in improvements. One year leaped from 79.8% accuracy to 94.3%.

AIDOC: What are your final thoughts on AI and its impact on the future?

Dr. Atin Saha: I think that it is helping people to triage and use it as a tool that works with radiologists to support their diagnoses. I think it has immense potential as a tool to help residents, medical students, or any physician, to train and learn better. With the ever-increasing volume of cases, it can analyze and present results that allow one to not only improve making the diagnosis but also aids in making it sooner which ultimately impacts patient care in a positive way. It will not replace the radiologist; it will enhance the work that they do.

Leave a comment

Recent Posts
Latest Tweet
See Demo Tailored for your PACS