A study published in European Radiology examined the prognostic efficacy of automatic segmentation of diffuse large B-cell lymphoma (DLBCL) in three-dimension (3D) FDG-PET scans using a deep-learning approach. The study authors, led byChong Jian, stated that their analysis validated their model’s ability to “accurately segment lymphoma lesions and allow fully automatic assessment of [total metabolic tumor volume (TMTV)] on PET scans for DLBCL patients.”
The report also noted that researchers verified predictive TMTV, or pTMTV as an independent prognostic factor for survival in patients with DLBCL. Performance metrics included Dice similarity coefficient (DSC), Haccard similarity coefficient (JSC), sensitivity, positive predictive value (PPV), Hausdorff Distance 95 (HD 95), and average symmetric surface distance (ASSD).
The deep learning–based prognostic model was constructed using two PET-CT datasets. The training cohort included 297 patients and the validation cohort included 116 patients. According to the report, the model’s “3D U-Net architecture was trained on patches randomly sampled within the PET images.” The authors calculated mean DSC, JSC, sensitivity, PPV, HD 95, and ASSD (with standard deviation), of 0.78 (± 0.25), 0.69 (± 0.26), 0.81 (± 0.27), 0.82 ± (0.25), 24.58 (± 35.18), and 4.46 (± 8.92), respectively.
Overall, the authors posited that their fully convolutional neural network model with a U-Net architecture was a feasible prognostic tool in the treatment of patients with DLBCL.