Season 4 Episode 2: AI in Cancer Research: Unlocking the Future of Healthcare

GOSH Podcast survey:
https://ubc.ca1.qualtrics.com/jfe/form/SV_9vJ2Jt3Mr6OPJBk

Description: In this episode, we delve into the dynamic world of cancer research, where artificial intelligence (AI) meets oncology to shape the future of healthcare. Our guest, Ali Khajegili Mirabadi, a PhD student in Bioinformatics at the University of British Columbia, takes us on a captivating journey through the impact of AI in cancer research and its potential to revolutionize the field. Discover how AI serves as a powerful ally in the battle against cancer, particularly rare and complex forms. Ali sheds light on the challenges of handling vast amounts of complex data and how AI helps identify patterns and similarities that may be imperceptible to the human eye. These insights have the potential to lead to groundbreaking discoveries in the world of cancer subtyping and personalized treatment.

Bio:  Ali Khajegili Mirabadi is a second-year PhD student in Bioinformatics at the University of British Columbia. His research focuses on Artificial Intelligence (AI)-based cancer patient similarity learning to advance clinical precision medicine. He primarily targets rare cancers across all organs, utilizing microscopic tissue images, clinical data, and genome information to enhance AI models' ability to identify patient similarities and trends in treatment. Supported by the UBC Four Year Fellowship in Dr. Bashashati’s research lab, Ali's work carries significant potential for improving patient care. Before joining UBC, he simultaneously achieved two undergraduate degrees in Electrical Engineering and Applied Mathematics with honors from the Isfahan University of Technology, Iran.

Link to the Dr. Ali Bashashati's lab website:
https://aimlab.ca/team/
_

For more information on the Gynecologic Cancer Initiative, please visit https://gynecancerinitiative.ca/ or email us at info@gynecancerinitiative.ca  
 
Where to learn more about us:  
 
Twitter – @GCI_Cluster 
 
Instagram – @gynecancerinitiative 
 
Facebook – facebook.com/gynecancerinitiative

Transcript

SPEAKERS 

Stephanie Lam, Ali Khajegili Mirabadi 

SUMMARY KEYWORDS   

Artificial Intelligence, Rare Cancers, Precision Medicine, Gynecological cancers, Research, Oncology, GOSH podcast 

Intro: Thanks for listening to the GOSH podcast. GOSH stands for the Gynecologic Oncology Sharing Hub, an open space for real and evidence-based discussions on gynecologic cancers. We'll share the stories of gyne cancer patients and survivors and hear from researchers and clinicians who are working behind the scenes to improve the lives of people with gynecologic cancers. Our podcast is produced and recorded on traditional unceded territories of the Musqueam, Squamish, and Tsleil-Waututh Nations. It is produced by the Gynecologic Cancer Initiative, a province-wide initiative in British Columbia with a mission to accelerate transformative research and translational practice on the prevention, detection, treatment, and survivorship of gynecologic cancers.   

Hi, I'm Nicole Keay, and I'm Stephanie Lam and you're listening to the GOSH, podcast. 

00:00:02 Stephanie 

OK, great. So welcome back to another episode of The Gosh Podcast. Today we have an exciting guest with us. Today we have Ali Khajegili Mirabadi. Ali is a second-year PhD student in Bioinformatics at the University of British Columbia. His research focuses on Artificial Intelligence (AI)-based cancer patient similarity learning to advance clinical precision medicine. He primarily targets rare cancers across all organs, utilizing microscopic tissue images, clinical data, and genome information to enhance AI models' ability to identify patient similarities and trends in treatment. Supported by the UBC Four Year Fellowship in Dr. Bashashati’s research lab, Ali's work carries significant potential for improving patient care. Before joining UBC, he simultaneously achieved two undergraduate degrees in Electrical Engineering and Applied Mathematics with honors from the Isfahan University of Technology, Iran. 

Thank you so much for joining us today, Ali. OK, great. So thank you so much for joining us on the podcast today. Ali, can you start off just by telling us a little bit about yourself? 

00:01:35 Ali 

Yeah, sure. It's a great pleasure to be here and talk about my work. And thanks for inviting me. So as you introduce, I'm Ali. I'm from Iran and I right now I'm studying by informatics, my PhD study at UBC. And I moved in Vancouver, July 2022. To start my PhD and yeah, everything I like about I like everything about Vancouver and yeah, here I am to talk about my work. 

00:02:07 Stephanie 

Great. Well, welcome to Vancouver. How are you enjoying the beautiful, sunny weather we've gotten this year so far? 

00:02:15 Ali 

Yeah, I love it. Yeah, summers in Vancouver is really nice going for a hike or each days. I enjoy them so I really like it. It's been a really good year for me. 

00:02:28 Stephanie 

I'm glad. Well, thank you so much for joining us on the podcast today. For my next question to you, I was just a little bit curious about kind of how you got interested in doing biomedical research and more specifically Cancer Research given your background in engineering and mathematics. So can you tell us a little bit more about how you got into this, uh, particular field. 

00:02:52 Ali 

Ohh for sure, that's a really good question. So it's a long story, but yeah, Long story short is that I kind of was interested in biology from high school, but the system in our country is that like we have a really strong computational feeling. Our country. So I decided to 1st for my undergrad study mathematics and electrical engineering to learn tools like because mathematics and engineering is all about tools and applications. But still I had like that interest in biology and discovering a thing. So after during my undergrad I was taking like Life Sciences courses and reading different articles about biology or cancer and a topic about pathogens was really interesting to me, so I started to read about pathogens and pathology and I found a field really interesting. So I think it was about four years ago, so I decided to continue that direction to combine my mathematics and engineering background in a way that I can use it for pathology, and I did some internal research internship like in the same in the similar field but different but it was about like stomach cancers and then I decided to apply for UBC bioinformatics program, which is in Doctor Ali Bashashati I contacted with them and it's really good research because they are working on cancers and working on cancer has a lot of how they say advantage because it's really complicated from science part you can learn a lot of stuff by using those mathematical tools you can discover a lot of new stuff, but at the same time it's really hard. So you need to spend a lot of time working with formulas or models to learn something and discover something, but the on the other hand which is I think, more meaningful part is that by working in biomedical science or Cancer Research, your work would impact people lives. So that's that kind of given meaning to my work. And I like it because of that. So I decided to do more research in cancers and it's like a interdisciplinary field. 

00:05:25 Stephanie 

Oh, that's great. Thank you for sharing that. It sounds like it was a really kind of nice marrying of the two kind of areas of interest that you had from a computational background, but then also your biology background and I think like you said, I think cancer is such a wide open field and there's still so much that we don't know about it that I think there's definitely a lot of really kind of interesting things for you to look at, which kind of makes me curious to learn a little bit about your research work. So could you share with the listeners today just a little bit about your research and what you're doing? Within your PhD degree right now. 

00:06:06 Ali 

Yeah, definitely. So Speaking of cancers, we know that cancers are really complicated and complex disease, so that we don't have even treatment or we can barely diagnose many of them because we have, like, rare cancers or some unknown cancers. So working with these cancers need strong tools. So in my research, mostly focusing on rare cancers, but my tool here is AI. So AI models and I think everyone nowadays knows about AI because of chat GPT and other like famous AI models. But they are mainly working on general feels like general, we call it general computer vision. But the story in cancer is a little bit different because as we said, the data for cancer is like more complicated, for example, like a simple human being can look that photo of a dog and cat and distinguished between them. But for cancer, it's not like that. When you have two different microscopic photos from cancers, one like someone some, like a person who don't have any education in pathology, cannot like easily differentiate between them. Because everything looks similar, so here it comes that we want to use AI. Because like for humans to learn and distinguish this pattern, it takes like maybe 5 to 10 years of fellowship or different residency programs to learn and to be an expert in the field. So our question was that, OK, can we use an AI model these strong, these strong tools to use them or maybe to help pathologists in their decision making. This process easier or not. So in my research work is. This is exactly the question so that we want to use AI models to kind of finds similarities in rare cancers and by rare cancers when we speak of rare, it means that we don't have enough samples, like maybe two or three in 10 years in Vancouver. So if you have two cases in 10 years in VGH (Vancouver General Hospital), so even pathologists don't have any broad. Knowledge about them. So we want to devise and model that can use them, and the idea is that OK, we have like maybe 10 different rare cancers. Let's go to our. Database and find those similar cases. And these similar cases can be When you find the similar cases, you can plan for treatment or define some strategies for your new patients to rely based on that, like for example, drug A has worked for patient B that we had in the archive. So let's try it again for this new patient. Some ideas like that, yeah. 

00:09:03 Stephanie 

Hmm, I see. And so for our listeners who aren't quite super familiar with AI, why is it so important that you have more than a just a couple of cases of I guess data for you guys to use and kind of how does that build into the model that you guys are creating? 

00:09:25 Ali 

Oh yeah, that's a really good question. So basically, when we talk about AI, we also talk about learning. So by learning then we have an understanding of learning. So when a child go to a high school, it started to read and practice to learn something. So this is true for AI as well. So when we have AI model, we need to teach it something by teaching, we need to have data and this process of teaching can be expensive and by being expensive I mean that AI is not as efficient as humans right now we are trying to make them efficient in a way that we as few sample as possible that we have, we can teach them. But right now, it's not like that. So we need a lot of samples from a data to teach them to tell them, OK, learn this pattern from the data. So for example like. Uh, when you have two samples from a cancer, you cannot teach something, but when you have 200 cases. Yeah you can do something. 

00:10:34 Stephanie 

I see. So really the data is utilized to teach artificial intelligence model. How to accurately kind of make a decision or to tell the difference between the different diagnosis that are the rare cancers. Am I hearing that right? 

00:10:55 Ali 

Yeah. Yeah, you're right. Yeah. The end goal is to predict cancer or to find a pattern in the data. This is the end goal that the idea is the same that the doctor wants to find the diagnosis to see. OK, this is a subtype from cancer A. So this is the same thing that AI is doing.  And by learning the model is doing your search in the data to find something and make a decision based on that. 

00:11:26 Stephanie 

Yeah. Ohh very interesting. So I'm curious. You know, I think we talked a lot about kind of helping to make diagnoses and with rare cancers and stuff like that with the work that you're doing right now, how have you and the team started to think about how this would impact the healthcare system today or you know a couple of years down the road, how would it be applied in a real life setting? 

00:11:55 Ali 

Yeah, that's a great question. So, I think it will affect every aspect of our current healthcare system because AI can be utilized in different parts. For example, from a screening to treatment or to risk identification. And it can be used to make the workflow that currently we have more efficient and more streamlined. So for example, when our pathology lab are working with cancer samples they take many samples from the patients. But a lot of these samples don't have any useful information for the pathologist, but the pathologies should sit behind microscope and test and check every single sample to find the one with tumor. But with having an eye. You can easily remove those samples without any useful information and pass the samples with content tumor to the pathologies. So this is a very simple use case that can make the process very fast for the pathologies. So pathologies Instead of spending hours on investigating useless samples can spend a few minutes on a useful case and find the other use case is that our team has worked on and they got a really good result is that in the endometrial cancer. They were working and they were using this AI models and they saw at some point AI is seeing something that human eyes cannot see and that was that AI understood that there is a group of patients that used to be considered as a subtle cancer called NSNP, non-specific molecular profile. But they said that OK, this group of patients have two subgroups. And it was like very interesting that why is it like that? So they started to define the hypothesis that they are actually two different subtypes. And they evaluated that hypothesis and tested that, and they proved that yes, they those are two different subtypes within that previous subtype. So that was a new subtype discovery which was available by AI. So that's something that makes us able to do it, you know. Why? Because the data is so complex. Our human eye is not able to see all the patterns. But because AI is more accurate and less bias. It's not accurate to say less bias, but yeah, in very roughly it's less bias to identify some certain patterns. We can do those studies as well and by finding a new subtab, you basically identify a new group of patients that you can design new treatment or a strategy for targeted therapy, even for them. So kind of helping them to get better treatment. 

00:15:07 Stephanie 

Wow, that's really amazing. I'm hearing that it's both kind of a from an efficiency perspective and trying to make the lives and kind of system run more efficiently. So then the system can see more patients, but also that there is an opportunity to really utilize AI to, you know be more specific in the cancer treatment that patients can receive and hopefully that improve patient outcomes. At the end of the day. Which is crazy to kind of think that AI is response can do all of those different things. You know, we talked a little bit about some of the challenges working with AI and obviously I think kind of access to that initial data is really huge in your role as a PhD student right now and working in this field. Have you encountered any other challenges when it comes to working with AI and in Cancer Research? 

00:16:08 Ali 

Speaking of challenges, we have different types of challenges. Here, like challenges that are directly associated with the data and cancer challenges that are directly associated with the AI itself, or challenges that we have with healthcare system. So with the data is the, if I want to mention a few of them, is that the data is really complex and is really big. For example, one cancer tissue image is around 1000 times bigger than one of our simple iPhone photos that we take every day. Wow. So it's like 5 GB image. One single image of its small tissue is 5 gigabytes, so processing that is really hard, so that's a challenge for that comes from the data. But the challenge that comes for AI itself is that AI is not accountable, meaning that a decision that is made by AI. There is no one behind it to say, OK, I'm responsible for that decision or in healthcare system, if your AI system makes a wrong decision, who's responsible for that? So that's a challenge for AI or the other challenge is that AI systems right now are not fully interpretable. Meaning that we don't know what is going on exactly in the model that the model will predict something. We are trying to improve modeling to make it fully explainable. And there are, like, challenges with hospitals is that different Hospitals have different sharing protocols, so that makes gathering data way harder. So it's like a challenge that OK, hospitals don't share their data. So we have less data right now. So we cannot like have a better AI system. 

00:18:09 Stephanie 

 Yeah. Clearly, there's a lot of challenges that you and your team were trying to work through it. I think some of the ones that are really interesting to me is just kind of from both a kind of a data and development and perspective, but then also how it applies into the real world. And how it how you know we develop these technologies but also we to think about what we need to put into place into our systems when we are trying to roll it into our a real life environment like you said. You know, there's no one accountable to it. So how do we build some structures to make sure there is the accountability or whatever there needs to be so that it could be actually used effectively. Yeah, very interesting. 

00:18:56 Ali 

And that's totally true. Yeah. Right now, the idea is that for healthcare, AI can be used as an adjoint tool or as an assistant tool that make the process faster or more optimal. 

00:19:13 Stephanie 

Yeah, yeah, which I think is already a huge, really important kind of tool that is really needed in our healthcare system with, you know, trying to make things more efficient is always a goal within the healthcare system and making sure that we're leveraging, you know, the people to the best of their ability. And you know like pathologists making sure that they're spending their time looking at the really kind of important ones instead of just kind of going through slides and slides of kind of things that might not really help patients at the end of the day right?  So I think it's definitely a tool that's is can be really useful in our healthcare environment. OK, so the very last question that I have for you is just around kind of what your thoughts are on the future of AI and healthcare and kind of what your you are most excited about in this space? 

00:20:12 Ali 

Ohh yes, for looking ahead if I want to describe some of the like impacts that AI can have on healthcare system is that I think it would be more streamlined. It's not like that, “OK, AI will replace everyone or Doctors will be replaced”. It is not like that. Because I said a computer system is never accountable for something for decisions. So we always need the doctors to supervise any decision that is being made. But I think that in close future we will have these AI systems that are implemented in our labs, in our hospitals that our diagnosis would be very fast. They help us from sample preparation to treatment plan, to defining different treatment plans for each cases. I think we will have a day that each patient, each cancer patient will have a specifically design treatment for themselves based on their genome based on their data. And I think this will happen thanks to AI. I think because we can process this big amount of data and with the help of AI, I think because people are right now working on some methods that they can even design drugs using AI. So they try different molecular design and test them in the lab, which is like they using AI in this part is like making the process faster. And basically more accurate. So why not we use AI for simulating everything. So that's really helpful. But the exciting part for me is that like I said we want something that we can use it in the clinic. Right? If we have a really big model, but it's not useful for the clinic, it doesn't have any value. We design a system AI system that can be used in the clinic right now and our team is working on the models, but at the same time we are working on a platform to share it with doctors which pathologies to make them to be able to use AI system with in different computational tools, different area systems, and also statistical system to evaluate the decision. So this is exciting for me because right now we have like a lot of useful AI tools that we can use them for cancers. 

00:22:42 Stephanie 

Oh, oh, wow. There's definitely a lot going on in this space. So it’s really exciting to hear about what's going on and kind of what the future looks like. I think the technology is definitely, you know, new and exciting, especially in the healthcare environment. So really looking forward to kind of hearing more about your work as you move along your PhD journey and hopefully we can have you back on the podcast in a couple of years time when you're further along in your journey and can hear more about your work at that point in time. But thank you so much Ali, for joining us today on the gosh podcast. And we'll be sure to link in the show notes some uh links to your lab, and those some of the work that you're doing right now. So thank you for taking the time today. 

00:23:36 Ali 

Thank you so much for having me. Yeah, it was a really nice talk. Nice talking to you. And yeah, it's a really good effort to spread these to patients and helping them to know more about AI and our work. And yeah, thank you so much. And if anyone has any questions, they can reach out to us they will happily respond any question. Thank you so much.