Introduction

Thank you. Thank you, everyone. And very good evening to everyone.

It's my pleasure and, in fact, honor to be here and sharing the stage with MindStone. Thanks, Josh, for this opportunity.

I have actually a unique journey. I'm not a computer scientist or a code writer or a scripter, but I'm a clinical research scientist.

1What research we do in our lab, it takes decades to translate. What does it mean? It means the findings we do now, it will take decades to translate for the treatment purpose.

And it's a huge gap. Unfortunately, many of the researchers, they don't see their findings to translate it for the real world.

So with this AI and this technology, it's really an opportunity to leverage that and bring this gap.

So that's my kind of passion. How I can... close this gap of decades to translate.

My Journey and Passion

My specialty is in lung infection and inflammation, but as you saw in the previous Josh's introduction, now this is the age of AI. How we can leverage AI to minimize all the hurdles and bring what we're doing innovation to the purpose of translation for the community.

So I'm going to specifically highlight one of the use case of that is lung diseases.

We have seen the catastrophic effect of pandemic, but now we are happy this pandemic is gone. It's not in fact, we are still struggling with many, many mortalities with lung problems.

Challenges in Healthcare Data

But coming to that, I'm just quickly very, very high level go through the data because especially in healthcare, getting data is problem.

Even if you think for one disease and you want to solve the problem, the data is not one. It has a multi-model, very diverse imaging. Imaging could be MRI, X-ray, PET scan, ultrasound, doctor's report, the treatment, the EHR.

The data governance and data orchestration is very, very important. Otherwise, the saying would be right here, garbage in, garbage out.

So this is just a high level. We need to focus on the data governance before even going to the next step of, especially for the healthcare.

This is my personal experience.

Role of AI in Healthcare

And differently, AI, now AI is like everywhere. Sometimes we use AI for the sake of AI. So we have to understand why we are using AI and how we're going to use AI and what could be the value proposition, how that AI is going to help or bring the value.

So these are the framework.

Based on these two factors, we work on a lung anomalies detection. See, this is that I just started.

Focus on Lung Diseases

We are happy that there is no more pandemic, but there are around 6 million deaths happen every year because of the lung anomalies, lung diseases. And if you further deep down, those lung anomalies are only like three diseases, lung cancer, COPD, pneumonia, and tuberculosis. Because of only these three lung problems, most of the mortalities are happening.

So now the next question is, why we have technology? We have done so much advancement in health care. Why this much that's happening every year? There are reasons.

Traditional Workflow

So this is a workflow. It's a traditional workflow, how we diagnose the problem, how we treat, and how we provide the treatment. And that takes a substantial amount of time from diagnosis to treatment to their effect.

This is definitely that needs expertise. Along with that, this is really expensive. And the bottom line is very challenging, very disheartening.

Like, even in North America, I'm working in sick hospital, there are 20 to 25% misdiagnosis happening. Misdiagnosis means what? Misdiagnosis means death.

We are not treating, rather we are neglecting. We are not able to diagnose a disease correctly. And especially lung cancer misdiagnosed as COPD.

Unfortunately, if someone has lung cancer and he or she diagnosed as something else, patient is happiest and just at home. And every day is like devastating for him or her.

The second problem is we do have medical expertise. A radiologist or a pulmonologist, he or she has years of experience, but that experience is respected. But as a human, experience is experience. That experience is very, very narrow.

Reducing Challenges with AI

So with the AI, this we can minimize.

Now I'm coming to the next one, how we can reduce these challenges. So even if we provide a very simple, as a layman language, a diagnostic tool which is trained with this multimodal data, the X-ray, HRCT, MRI, patient record, and so on, that will definitely minimize the misdiagnosis. Because the output of AI is not based on one report or two reports. That could be based on the entire data set which we have trained. This will definitely minimize the misdiagnosis.

It will help to triage the patient's specialty. Let's say if a patient has just pneumonia or flu, that patient should be taken in a different way. A patient who has some critical problem, whether it's COPD, lung cancer, fibrosis, that needs the special care.

So this will help to minimize... the patient load in our hospitals, in our emergencies, along with that, the patient will get a proper treatment. Definitely, at the end, this will bring a better health care output.

Developing a Diagnostic Tool

So based on that need, we developed a hypothesis and idea to bring radiologists, clinicians, and data scientists and data engineers to build a tool that helps to diagnose by taking the patients imaging data, their symptoms, geographical problems and genetic data to provide precision diagnostics. So this tool is going to be helpful in terms of minimizing misdiagnosis, providing a better care and triage and minimizing the load in our hospital's wait time.

So this is a very simple layout how this tool has worked differently.

I'm supposed to showcase the demo. I'll try to showcase the demo in the next one.

Challenges in Training Models

And here, the challenge is lung diseases are different. Every disease is different to others. COPD is different than lung cancer. Tuberculosis is different than the fibrosis, and so on.

So the training, one model Getting data for all the lung diseases and building one model is really, really difficult because the accuracy we are not getting that much.

So what we did here, we trained the model based on each diseases. Let's say pneumonia, fibrosis, sepsis, lung cancer. So that's one phase.

In the second phase and in the next layer, all the models were unified and the output is there as we can see in this structural diagram. And this is another, deep down, how this tool is working.

Tool Workflow

So let's say there is a patient, unknown patient. We don't know what problem that patient has.

The tool is going to only classify normal or abnormal. Abnormal means anomalies. And if there's anomalies, what are the key anomalies that are leading to the major mortalities?

And if those anomalies are lung cancer, COPD, pneumonia, or tuberculosis, so that is a priority. Especially if someone is diagnosed with lung cancer, then their further diagnosis and clinical intervention will be done.

Focus on Lung Cancer

Why we are doing here in the lung cancer? The second problem in case of lung cancer, and most of the deaths happening because of lung cancer in lung anomalies, most often when...

Oncologist prescribed chemotherapy. Sometimes chemotherapy won't respond. And patient comes after one or two or three cycles and there's no response.

Then the oncologist prescribes different sets of chemotherapy. But see at that point of time how much suffering patient has gone through. And maybe at the next level, patients don't have the strength to tolerate the chemotherapy, along with the expenses.

Integrating Genomics and Biomarkers

So based on now here, we are integrating genomics and biomarkers data of lung patient, the patient who has been diagnosed as lung cancer, so that oncologists get the information about the mechanism of the lung cancer. And then he can prescribe the chemotherapy, which will respond. So here, we're going to bring up personalized care.

So this is overall workflow, how it works. In demo, I am going to showcase up to here, because that part is still not there.

And again, apologies, Josh. I'm supposed to showcase the real demo here, but because of some technical issues, and I'm not getting connectivity to my server, but fortunately I have a video that will kind of show a demo.

Demo and Future Steps

Breathe easy. Feel secure.

Meet Palm Modeep. Palm Modeep is your personal health companion, designed to predict lung diseases early and accurately using advanced AI technology.

It collects anonymized patient data, analyzes it with advanced AI, and delivers detailed disease reports and personalized recommendations through a user-friendly interface. Palmodeep is designed to detect and monitor a range of pulmonary conditions.

It is easy to use. Let us walk through how it works.

First, go to the homepage and select the prediction model you need. Browse your local files and select the X-ray image you want to analyze. Then, simply upload the image.

AI algorithm analyzes the image to provide you with accurate results. You'll receive a detailed report with in-depth analysis.

Visit Digibiome.

Current Status and Collaborations

So before this one, this MVP is ready and I'm in discussion with Sickles Hospital and Trillium Hospital where we are in a process to get the retrospective data. So the data like last year or even before to validate this and after this, validation of this device by retrospective data, next phase is to integrate in their workflow.

And then we can start getting the output of this device in a real world scenario. So these are the two steps we are working on it.

And I have a huge collaboration from clinical sites. along with the data scientists, and I again say this is happening because of the AI.

Conclusion

So I made it at a very high level, not too much technical, but definitely if you have any questions, suggestions, please feel free to add here. Thank you.