Discriminating pseudoprogression and true progression in diffuse infiltrating glioma using multi-parametric MRI data through deep learning

Differentiating pseudoprogression from true tumor progression in diffuse infiltrating gliomas is challenging, potentially delaying treatment. This study used multiparametric MRI data as input for a CNN-LSTM deep learning model to discriminate between these conditions. Using 43 biopsy-proven diffuse infiltrating glioma cases, the CNN-LSTM model, incorporating multiple MRI sequences, outperformed a VGG16 model using single sequences, achieving higher accuracy and AUC. The findings suggest that using a CNN-LSTM model with multiparametric MRI data improves the diagnostic performance for differentiating pseudoprogression and true tumor progression.

Joonsang Lee, Nicholas Wang, Sevcan Turk, Shariq Mohammed, Remy Lobo, John Kim, Eric Liao, Sandra Camelo-Piragua, Michelle Kim, Larry Junck, Jayapalli Bapuraj, Ashok Srinivasan, Arvind Rao