Classification of Pediatric Posterior Fossa Tumors Using Convolutional Neural Network and Tabular Data

Abstract

Posterior fossa tumors (PFT) are the most common tumors in children. Differentiation between the various PFT types is critical, as different tumors have diverse treatment approaches. This study proposes the use of fused architecture comprising two neural networks, a pre-trained ResNet-50 Convolutional Neural Network (CNN) and a tabular based network for the classification of PFT. The study included data for 158 MRI scans of 22 healthy controls and 136 pediatric patients with newly diagnosed PFT (63 Pilocytic Astrocytoma, 57 Medulloblastoma and 16 Ependymoma). The input data for classification were from magnetic resonance imaging: post contrast T1-weighted, fluid attenuated inversion recovery and diffusion Trace images, and tabular data: subject's age. Evaluation of the model was performed in a stratified 5-fold cross-validation manner, based on accuracy, precision, recall and F1 score metrics. Model explanation was performed in terms of visual explanation of the CNN by Gradient-weighted Class Activation Mapping (Grad-CAM) and by testing the contribution to the classification results of the different imaging input data sets and the proposed fused architectures relative to CNN only and tabular only architectures. The best classification results were obtained with the fused CNN + tabular data architecture, and based on diffusion Trace images, achieving mean cross-validation accuracy of 0.88±0.04 for the validation and 0.87±0.02 for the test dataset. Overall, the proposed architecture achieved improvement in accuracy and F1 score compared to CNN method for this dataset.

Pediatric brain tumor classification using deep learning on MR-images from the children’s brain tumor network

“…Moreover, even though MCC and accuracy values for models trained on T2-w data were slightly lower compared to when ADC data was used, class-wise metrics for EP were higher in the case of T2-w explained by the larger number of images. The results on ADC data were in accordance with those of Artzi et al, 12 . However, in our study, only 1383 ADC images were available with a high class imbalance and with EPs presenting only seven scans to split between training and testing.…”

“…It should be noted that the comparison of the performance metrics among the studies should be considered in general terms due to the differences in the datasets and the number of classes for the defined tasks. The reliability of the results is also affected by the experimental protocols, as Quon et al, 11 applied a five-fold stratified cross-validation and employed the top five models and Artzi et al, 12 implemented a five-fold stratified cross-validation to compute the final predictions. The work presented here, unlike the two papers mentioned, employed the entire dataset and tested the network on all subjects in the dataset by implementing five repetitions of a ten-fold stratified cross-validation to achieve the most consistent and reliable results possible.…”

Convolutional Neural Network Techniques for Brain Tumor Classification (from 2015 to 2022): Review, Challenges, and Future Perspectives

“…Even though CNNs have demonstrated remarkable performance in brain tumor classification tasks in the majority of the reviewed studies, their level of trustworthiness and transparency must be evaluated in a clinic context. Of the included articles, only two studies, conducted by Artzi et al [ 122 ] and Gaur et al [ 127 ], investigated the Black-Box nature of CNN models for brain tumor classification to ensure that the model is looking in the correct place rather than at noise or unrelated artifacts.…”

“…The authors in [ 122 ] proposed a pre-trained ResNet-50 CNN architecture to classify three posterior fossa tumors from a private dataset and explained the classification decision by using gradient-weighted class activation mapping (Grad-CAM). The dataset consisted of 158 MRI scans of 22 healthy controls and 63 PA, 57 MB, and 16 EP patients.…”

“…Beyond showing accuracy gains, the surveyed articles rarely examined their generalization capability and interpretability. Only very few studies [ 73 , 120 ] tested their classification models on an independent dataset, and only one study [ 122 ] investigated the Black-Box characteristic of CNN models for brain tumor classification to ensure that the model they obtained was looking in the correct place for decision-making rather than at noise or unrelated artifacts.…”