My journey in data science

With a decade of experience solving real-world problems using machine learning, neural networks, time series forecasting, and NLP, Viraj Kulkarni leads the data science team from DeepTek to create deep learning models for radiology. Having co-founded CPC Analytics and Algocraft and published various research articles on radiography and deep learning models, Viraj adds another feather to his track record by pursuing a doctorate focused on the implementation of quantum computing in the AI.

Analytics India Magazine connects with Viraj as he recounts his dynamic journey through data science, machine learning and AI models.

AIM: How did you get started on this data science journey? Where did it all start?

Viraj: I delved into pattern recognition-based topics during my undergraduate studies, but my journey began during my post-graduate studies at Berkeley, where I started developing algorithms for NLP, understanding text, interpreting it and extracting feelings and opinions from it in 2011. Back Next, AI, NLP ML and data science were not widely used terms. And so, after I finished graduate school, I came back to India and started a company called Algocraft, where I continued the work I started in Berkeley. Algocraft relayed my next company, CPC Analytics, a consulting firm based in India, Germany and France, working with Fortune 500 companies and government organizations to set up their data strategy.

AIM: How important is it for aspirants to start early?

Viraj: Data science is quite a broad field. Machine learning models for making predictions, analytics dashboards for extracting insights from data, data pipelines for working with data – data science encompasses many different activities. There are also many intersections between data science and domain expertise. Building a portfolio early on is important in some of the areas above, but overall I would say no, you don’t need to start very early. There is no age limit or pressing need to start early. I know some great data scientists who started quite late in their careers. Starting late, in fact, gives you a very clear advantage. Work experience in other industries helps you develop a solid understanding of this field, which then enables you to build better data science tools. The problem with starting too early is that aspirants only focus on data science concepts and end up ignoring other aspects like model development, software engineering, or domain expertise. The lack of these skills makes him a superficial data scientist. And that’s certainly not what the industry or the world needs.

AIM: You have been featured in many research papers based on ML models, neural networks, etc. Is there an item that stands out?

Viraj: I recently wrote an article about the key considerations machine learning developers need to take into account when moving your AI models from the lab to hospitals. Although this article aims to develop AI in healthcare, these key considerations are generally applicable to all sectors. This article stands out because it addresses a very critical challenge. Developing AI in the lab is easy, but getting it from the lab to real hospitals or the place of deployment, where it creates value in practice, is a long journey. This article discusses challenges such as insufficient training data, imbalance in class representations, model generalizability, model degradation and dataset drift, etc. The article talks about those factors you need to consider to develop AI that is robust, accurate, usable, and valuable in practice.

AIM: How is DeepTek leveraging AI to help radiologists deliver better and faster reports?

Viraj: From a single chest X-ray, a radiologist can identify more than 60 different abnormalities. Many of these results are very difficult to spot or can easily be missed. Most radiologists have a high workload and work long shifts. This leads to human fatigue leading to more errors. The AI ​​we are building at DeepTek serves an important purpose, which is to direct radiologists’ attention to abnormal parts of the exam. The AI ​​detects that part of the analysis seems abnormal and highlights it. And then in addition to saying that this region looks abnormal, it also suggests different pathologies or findings that might be present in this region. With this solution, you quickly look at the scan, which is pre-read by the AI, you look at the regions that have been identified as abnormal, you click to approve or reject the AI’s predictions, and the approved predictions are automatically added to the report. So, AI helps in three ways. First, it improves reporting accuracy allowing radiologists to identify more abnormal cases. Second, it saves time reading each scan. Along with this, it also improves the standard and quality of reports.

AIM: What kind of milestone do you want to reach in the future?

Viraj: The adoption of AI in healthcare is growing very rapidly and the landscape is changing. I think the ultimate goal is to develop an AI that can autonomously read and report scans. This is the final destination that we would like to reach. Of course, the journey to this point is not straightforward. We recently authored another research paper that discusses the evolution of AI adoption in healthcare. We compare it to stages of autonomous driving from levels 0 to 5. We want AI in healthcare to achieve this level 5 of autonomy. Different research groups and companies are taking different routes to this important milestone. At DeepTek, we believe this path will be a winding one. To achieve this, we will first have to overcome many other challenges.

AIM: Will AI replace medical personnel in the future?

Viraj: This is an interesting point. AI will not replace doctors. We don’t see that happening. But what we see happening is that doctors who use AI will, over time, replace doctors who don’t use AI. We want doctors and healthcare professionals to see AI not as a threat, which I don’t think, but rather as an intelligent assistant that helps them do their job better and faster, and ultimately account, to improve patient outcomes.

AIM: What kind of change do you envision in the world of AI over the next decade?Viraj: AI is everywhere. It has already permeated all industries to varying degrees. But, we are still trying to understand the effects that AI will have in the long term. An important trend I see is an increasing focus on developing responsible AI that does not discriminate against any group and shows no bias. Instead of just providing an output, can the AI ​​put that output into the right context and give enough information to explain where the outputs are coming from? Often, we see that at the time of deployment, the AI ​​works very well, but then it degrades over time. Constantly monitoring this drop in performance is crucial to driving the adoption of AI across industries. Over the next two years, I think the focus will be on explainability, fairness, ethics, and accountability demonstrated by AI. I think the community at large as well as the AI ​​developer community needs to take this very seriously.

Sean N. Ayres