The use of convolutional neural networks (CNN) to analyze maximum intensity projection (MIP) images from positron emission tomography (PET) baseline scans from patients with diffuse large B cell lymphoma (DLBCL), could improve the prediction of treatment response, according to a recently published study in Scientific Reports.

Currently, R-CHOP chemotherapy represents the standard of care for patients with DLBCl. The decision between using R-CHOP as initial therapy and using more intensive regimes often depends on factors such as the disease stage at the moment of diagnosis and the score in the international prognostic index (IPI).

Read more about DLBCL prognosis

Despite these prognostic tools, a significant percentage of patients are resistant to standard-of-care therapy or relapse after initial remission, and patients with resistance to treatment or a history of relapse have a poor prognosis, the researchers noted.

“Therefore, early identification of high-risk patients is important as patients might benefit from a more tailored treatment strategy,” the authors wrote.

F-fluorodeoxyglucose (18F-FDG) PET scan is an integral part of the IPI score and allows the quantification of objective parameters such as metabolic tumor volume (MTV). However, these measures are observer-dependent. The authors hypothesize that CNN could replace the observer in this task improving the predictive capacity for response to treatment.

The study used 18F-FDG PET scans collected from a clinical trial including over 250 patients to compare the performance of CNN and IPI in predicting the probability of time to progression (TTP). The authors used an external data set from another clinical trial, including over 300 patients, to validate the results.

The CNN model had a better performance than the IPI model in predicting TPP, with the first one having an area under the curve of 0.74 compared with 0.64 of the latter.

“Even though further investigations are necessary, our current findings suggest that CNNs using MIPs have potential as outcome prediction models,” the authors concluded.


Ferrández MC, Golla SSV, Eertink JJ, et al. An artificial intelligence method using FDG PET to predict treatment outcome in diffuse large B cell lymphoma patients. Sci Rep. Published online August 13, 2023.