AUTHOR=Attallah Omneya 

TITLE=CoMB-Deep: Composite Deep Learning-Based Pipeline for Classifying Childhood Medulloblastoma and Its Classes

JOURNAL=Frontiers in Neuroinformatics

VOLUME=Volume 15 - 2021

YEAR=2021

URL=https://www.frontiersin.org/journals/neuroinformatics/articles/10.3389/fninf.2021.663592

DOI=10.3389/fninf.2021.663592

ISSN=1662-5196

ABSTRACT=Childhood medulloblastoma (MB) is a threatening malignant tumor affecting children all over the globe. It is the foremost common pediatric brain tumor causing death. The early and accurate classification of childhood MB and its classes is of great importance to help doctors choose the suitable treatment and observation plan, avoid tumor progression, and lower death rates. The current gold standard for diagnosing MB is histopathology of biopsy samples. However, analyzing histopathological images manually is complicated, expensive, time-consuming, and highly dependent on pathologist expertise and skills which might cause inaccurate results in some cases. This study aims to introduce a reliable computer assistant pipeline called CoMB-Deep to automatically classify MB and its classes with high accuracy from histopathological images. The key challenge in this study is the lack of Childhood MB datasets especially its four classes and the inadequate related studies. All related works were based on either deep learning (DL) or textural analysis feature extractions. Such studies employed distinct features to accomplish classification. Besides, most of them only extracted spatial features. Nevertheless, CoMB-Deep blends the advantages of textural analysis feature extraction techniques and DL approaches. CoMB-Deep consists of a composite of DL techniques. Initially, it extracts spatial deep features from 10 CNNs. Then, it performs a feature fusion step using DWT which is a texture analysis method capable of reducing the dimension of fused features as well. Next, CoMB-Deep explores the best combination of fused features which enhances the performance of the classification process using two search strategies. Afterwards, it employs two feature selection techniques on the fused feature sets s. A bi-directional LSTM network is utilized for the classification phase.  The results of CoMB-Deep prove that it is reliable. The results also indicate the feature sets selected using both search strategies have enhanced the performance of LSTM compared to individual deep features. CoMB-Deep is compared to related studies to verify its competitiveness and this comparison confirmed its robustness and outperformance. Hence, CoMB-Deep can assist pathologists in performing accurate diagnoses, reduce risks of misdiagnosis that could occur with manual diagnosis, accelerate the classification procedure, and decrease the cost of diagnosis.