ITU News recently caught up with Shinjini Kundu, Research Fellow at the University of Pittsburgh Medical Center, to discuss her ‘Transport Based Morphometry’ technique, or TBM, which allows for insight into artificial intelligence (AI) reasoning in the processing of medical images.
Transport-based morphometry – or TBM – is a foundational AI technology that enhances the ability of doctors to see diseases that are otherwise imperceptible to the human eye. What is exciting about TBM is that it has the potential to find diseases early, even before obvious signs can be seen on medical images.
TBM is not just a tool for doctors, the most exciting implications are actually for patients. Many of my patients have osteoarthritis, where the thin cushion between two bones – called cartilage – falls apart and causes bones to painfully grind against each other. In fact, thirty million people in the United States have it, with many more worldwide. Currently, I can’t confirm the diagnosis until I see the bone damage, but treatment options are limited by this point.
Imagine if we didn’t have to wait until bone damage and pain to detect osteoarthritis. What if we could predict who would develop symptoms before they happened? We tested this idea using TBM.
On images of knee cartilage from healthy people, we used TBM to detect subtle patterns not seen by humans. TBM used machine learning to link these patterns to future osteoarthritis symptoms.
We were able to predict osteoarthritis symptoms in healthy people 3 years in the future, with up to 86% accuracy on patients whose scans TBM had never seen before. The diagnostic potential for this technology is huge. The idea that TBM can hasten disease detection is exciting and could empower many patients to take charge of their health early, before troubling symptoms develop.
From sequencing the individual genome to sophisticated biosensors, we have more data about health today than ever before. Artificial intelligence gives us a chance to derive actionable insights from the ocean of data that we are currently drowning in. The boon of AI technology is that it is highly scalable. Coupled with appropriate data availability, AI has the potential to touch all 7.5 billion people currently inhabiting the planet.
“The greatest benefit from AI will come when we go beyond boundaries. We need to collaborate across countries, societies, and industries.”
However, there are also several caveats. The first is that AI systems depend on data quality. Therefore, we need to first curate health data that are accurate, representative, and unbiased.
Second, the promise of scalability will come with the need for some sort of data standardization. For example, telephone standards across the globe help us communicate with each other. Similarly, if there were a global standard for imaging data acquisition, we could more directly compare these images from patients across the world. The third challenge is transparency and trust. Scalability of AI is a double-edged sword because mistakes can also scale.
Yes, I think we need transparency into AI systems. Especially in the medical realm, there are high stakes to our decisions. The partnership between human and machine intelligence needs to be built on a foundation of trust.
When an AI system provides an answer contrary to human intuition, it is important to understand its reasoning. We need to figure out if the AI system is trying to teach us something, or if it is simply making a mistake. Transparency also ensures that the AI systems reflect our human values of justice, autonomy, and respect that doctors give their patients.
I am a strong proponent of transparent AI, and my beliefs informed my approach to creating TBM.
Unlike traditional AI methods, TBM enables full transparency. When TBM has learned the hidden patterns enabling accurate detection of disease, it is able to generate realistic medical images that illustrate the hidden pattern that it has discovered. Therefore, we are able to map the reasoning in the TBM framework back to the level of cells, tissues, and internal organs.
I think there were three top takeaways from the AI for Good Global Summit. First, AI has immense potential, but it is up to us to make sure that it is equitable and empowers everyone equally. Second, the greatest benefit from AI will come when we go beyond boundaries. We need to collaborate across countries, societies, and industries.
Third, we need more community engagement because when individuals adapt existing technology in creative ways to solve local challenges, the results can be transformative.
I was an avid reader of anything that came my way as a child. What particularly kindled my imagination was science fiction stories. One of the writers that left an impression was Isaac Asimov, the writer of I, Robot. In his novel, he narrates the life and work of the fictional Susan Calvin, a “robopsychologist” – an expert in understanding the minds of artificially intelligent robots.
There has been much progress in the last few years in addressing the gender gap in tech fields, but we are not yet at parity.
Fortunately, I think there is momentum toward parity. In my own experience, the most valuable advice I had was to find mentors who are invested in your success. I pay it forward by being a mentor for young women interested in science, technology, engineering and mathematics (STEM) professions. I find it gratifying to show them that despite the obstacles, this is a rewarding field.