Purpose  

For the identification of high-risk patients in diffuse large B-cell lymphoma (DLBCL), we investigated the prognostic significance of in vivo radiomics derived from baseline [¹⁸F]FDG PET/CT and clinical parameters.

Patient’s Data

The baseline pretreatment [18F]FDG PET/CT scans of 85 patients diagnosed with DLBCL performed in the period between January 2014 and December 2019 were assessed. The [18F]FDG PET/CT scans were carried out in two centers: at University of Pécs, Department of Medical Imaging—Center 1 including 41 patients, and at University of Kaposvár, Hungary— Center 2 including 44 patients. The median age of patients in this study population was 59 years (range: 23–81 years) with 48.20% (n = 41) of patients older than 60. In this cohort, 40 (47%) patients were male, and 45 (53%) were female.  The patients with incomplete medical records and those who received nonstandard treatments were excluded from the final analysis.

Methods

Pretreatment whole-body [18F]FDG PET/CT scans were performed using a Mediso AnyScan® PET/CT scanner in 41 patients (Center 1) and a Siemens Biograph Truepoint 64 PET/ CT scanner in 44 patients (Center 2). All patients in the study were subjected to full history and complete clinical examination including the clinical stage of the disease. The patients were instructed to fast for 6 h before the scan. Blood glucose level was ensured to be below <8 mmol/L in all patients before the injection of radiotracer. Intravenous (i.v.) injection of [18F] FDG through an i.v. line with a dose of 3–4 MBq/kg was administered. After tracer injection, the patient was asked to stay for at least 60 min in a dark room covered by warm blankets. No speaking, chewing, or reading was allowed.

Two-year event-free survival (EFS) was defined. After delineation of lymphoma lesions, conventional PET parameters and in vivo radiomics were extracted. For 2-year EFS prognosis assessment, the Center 1 dataset was utilized as the training set and underwent automated machine learning analysis. The dataset of Center 2 was utilized as an independent test set to validate the established predictive model built by the dataset of Center 1.

Image processing

Lymphoma lesions were detected by InterView™ FUSION ver. 3.10 (Mediso Medical Imaging Systems Ltd., Budapest, Hungary) clinical evaluation software. The average SUV-max value of the liver (3.5–5.5) served as a reference threshold for the semi-automated algorithm. This approach was selected to minimize the effects of patient-specific radiotracer distributions. TLG, TMTV, and SUV-max were automatically calculated across all delineated lesions. Furthermore, SUV-peak values were segmented from the VOI with the highest activity. For further radiomic feature extraction, the largest VOI was selected in each patient.

Results

At the end of the standard induction therapy, 55 patients achieved complete metabolic remission. During the 2-year follow-up, 14 patients had primary refractory disease, 14 patients relapsed within 12 months, and 2 patients had relapsed within 24 months. In summary, after the end of therapy, 30 patients had detectable metabolically active tumor tissue and relapsed within 24 months (Figure 1).

Figure 1: Comparison of clinical outcomes based on maximum intensity projection (MIP) images in three patients (A–C). By each patient, the first image shows primary staging, the second shows interim PET scan, and the third shows post-treatment restaging scan. The red arrows indicate FDG avid lymphoma foci. (A) Patient in complete remission to treatment. The increased FDG uptake in all three images was a sign of thyroiditis. (B) Patient without complete remission during and after the therapy. The interim scan showed Deauville score 4. (C) Patient had an interim scan with Deauville score 3 but relapsed after the treatment.

The automated machine learning analysis of the Center 1 dataset revealed that the most important features for building 2-year EFS are as follows: max diameter, neighbor gray tone difference matrix (NGTDM) busyness, total lesion glycolysis, total metabolic tumor volume, and NGTDM coarseness. The predictive model built on the Center 1 dataset yielded 79% sensitivity, 83% specificity, 69% positive predictive value, 89% negative predictive value, and 0.85 AUC by evaluating the Center 2 dataset.

Conclusions

With our dual-center study, we could demonstrate that predicting 2-year progression-free survival in DLBCL patients is feasible with high-precision building on imaging and clinical parameters. This is in line with prior studies that utilize holistic datasets to build so-called holomics prediction models with machine learning . Given that our model yielded a balanced sensitivity and specificity, it could be a viable option to personalize the patient’s treatment. In the era of personalized medicine, with more detailed and specialized molecular diagnostics—especially in DLBCL—this could help clinicians to manage their patients more adequately and effectively.

Full article on Frontiers.