New, automated system calculates metabolic tumor volume in diffuse large B-cell lymphoma

Researchers at Sylvester Comprehensive Cancer Center at the University of Miami Miller School of Medicine have developed a new, automated way to calculate metabolic tumor volume (MTV) in diffuse large B-cell lymphoma (DLBCL). These findings could make it much easier to calculate tumor volume for clinical trials and possibly patient care. The study was published in Cancers.

"Metabolic tumor volume can have a profound impact on patient outcomes but, until now, there's not been an automated way to calculate it," said Juan Pablo Alderuccio, M.D., associate professor of medicine in the Division of Hematology and co-senior author on the study. "Now, we have developed an artificial intelligence-based system to get accurate tumor volumes both faster and with less human intervention."

Lymphoma patients with larger tumor volumes tend to have more challenging prognoses. As a result, clinicians have wanted to stratify patients based on MTV, but this has been quite difficult. Existing software is only semi-automated, meaning that radiologists have to pay close attention to the results and often fill in missing information, which can be a time-intensive process.

When we read a PET scan, there are both pathologic lesions and normal structures that light up. Radiologists review these images and differentiate between lesions and benign processes. Most software can't tell the difference between something that is benign, or has normal physiologic activity, versus a tumor."

Russ Kuker, M.D., associate professor of radiology in the Division of Nuclear Medicine and first author on the study

Faster, better readings

In the study, the researchers test drove a new, deep learning-based approach that fully automates MTV calculations. Dr. Kuker and colleagues used traditional, semi-automated methods to calculate MTV in 100 patients. Independently, medical physicist and co-senior author Fei Yang, Ph.D., in the Department of Radiation Oncology, reviewed the same scans using the deep learning algorithms.

"We found the results were highly correlated," said Dr. Alderuccio. "But even more importantly, instead of taking up to 30 minutes to come up with the answer for each scan, the machine learning approach took around five. By significantly decreasing the reading time, this opens the door to using MTV in clinical trials because you can use it on large numbers of patients and have the results right away."

This project was a multidisciplinary effort supported by Craig Moskowitz, M.D., Sylvester Comprehensive Cancer Center physician-in-chief, and initiated by Alan Pollack, M.D., Ph.D., who chairs the Department of Radiation Oncology.

MTV could be a valuable tool for clinical trials, stratifying patients by risk to better understand who responds to treatment and why. In addition, this technique might be used to advance clinical care, though it will need considerably more validation.

"We know that patients with higher MTV do worse," said Dr. Alderuccio. "Those patients may be more likely to receive intensified or experimental therapies to improve outcomes. Patients with low metabolic tumor volume might present better outcomes to standard therapies. This machine learning approach could give us better opportunities to make those distinctions."