In recent years, the low-dose CT scan has helped doctors better detect, understand and monitor lung cancer in patients. Using a series of images taken from different angles, this diagnostic test helps doctors create a more detailed internal view into a patient’s body.

Mohamad Abazeed, M.D., Ph.D., is particularly interested in how medical imaging like the CT scan can be used to not only detect lung cancer but also guide physicians toward more precise cancer therapies. As a physician scientist, he treats patients in addition to researching their diseases, making him positioned to champion a study that advances imaging that may improve treatment options. His Lung Cancer Discovery Award study, entitled "Mapping and Exploiting the Subclonal Architecture of Lung Adenocarcinoma," seeks to do just that.

After being awarded a second year of funding from the American Lung Association, we sat down with Dr. Abazeed to talk about what he has found so far and what he anticipates for the future.

What is Artificial Intelligence (AI)? The capability of a machine to imitate intelligent human behavior.

What is a neural network? Neural networks are a series of algorithms that mimic the operations of a human brain to recognize relationships between vast amounts of data.

So…Artificial neural networks (ANN) are: computing systems that are inspired by the biological neural networks that constitute human brains. Such systems "learn" to perform tasks by considering examples.

Q: Can you briefly explain the study in your own words?

A: We built an artificial intelligence (AI) machine that can help us identify tumor characteristics from a CT scan that can predict cancer treatment failures and guide these treatments. We built the AI framework using CT scans and health records from 944 lung cancer patients treated with stereotactic body radiotherapy. We input pre-treatment lung CT images into "Deep Profiler," a neural network that can pull large amounts of features out of radiologic images. The neural network analyzed and examined the scans to predict treatment outcomes. We then used the records from patients that the neural network was not initially privy to (for whom we knew whether treatments were successful or not, but the framework did not) to test its accuracy. Using this approach, we were able to accurately predict treatment failures across a range of clinical settings such as distinct tumor stages, CT scanners and treatment centers.  

Q: What makes this study one-of-a-kind?

A: To our knowledge, this study is the first to implement a neural network, or a technology that mimics the activity of the human brain to personalize radiotherapy. It has the ability to create new imaging features to predict the risk of failure for patients treated with radiotherapy (also called radiation therapy). Our framework could help doctors better predict which patients would benefit from radiotherapy and the ideal dose to fight the cancer without causing too many side effects.

Q: Why do you think this study is creating such a buzz in scientific community?

A: Patients with cancer present with mainly unchangeable variables: tumor type, stage, volume, image features, genetics, for example. Oncologists have within their purview one very important variable, however, and that is the type of treatment that is delivered and its intensity. Today, most patients receiving radiotherapy receive a generic dose based on the site of their disease.

Physicians and scientists have been clamoring to improve this one-size-fits-all approach for decades. However, the information and methodological frameworks have not caught up to this pressing need. This new framework represents the first opportunity to use imaging with medical record data to guide personalized radiation dose recommendations.

Q: What were you hoping to find and what were the actual results?

A: We were pleasantly surprised by how well the model worked. The model was able to consistently identify the tumors that were more likely to fail radiotherapy. Tumors are given a score by a computer based on how they look on the CT scan. Tumors that were expected to fail radiation therapy received a "high image" score and tumors that were expected to respond to radiotherapy were given a "low image" score. The AI machine predicted that tumors with a "high image" score were 4-5 times more likely to fail radiotherapy than those who had a "low image" score.

We confirmed the application of the method to other hospital settings by running an independent study of 95 patients who received lung radiotherapy at seven affiliate sites and were scanned using several types of CTs. The network was very accurate in sorting these patients into high- and low-risk groups.  

Q: With the renewal of your award, what are the next steps?

A: We will continue to advance this work by incorporating genetic markers with our image-based framework. We will also seek integration into the current physician workflow using software designed for the clinician. Our overall goal is to individualize cancer care by helping physicians recommend treatments based on the features of individual patients and/or their tumors.

Want to learn more about the American Lung Association Research Team? Read about the pioneering researchers and studies we're currently funding.

No upcoming events near you