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

Area of Science:

• Computational pathology
• Pediatric oncology
• Artificial intelligence in medicine

Background:

• Childhood medulloblastoma (MB) is a highly lethal pediatric brain tumor.
• Accurate MB classification is crucial for effective treatment and improved patient outcomes.
• Current histopathological diagnosis is labor-intensive, subjective, and prone to errors.

Purpose of the Study:

• To develop a reliable computer-assisted pipeline, CoMB-Deep, for automated MB classification from histopathological images.
• To address the challenge of limited childhood MB datasets and inadequate existing classification studies.
• To enhance diagnostic accuracy, efficiency, and reduce costs associated with MB diagnosis.

Main Methods:

• CoMB-Deep integrates deep learning (DL) and texture analysis for feature extraction.
• It utilizes 10 convolutional neural networks (CNNs) for spatial feature extraction, followed by discrete wavelet transform (DWT) for feature fusion and dimensionality reduction.
• The pipeline employs search and selection strategies to optimize fused features, using a bi-directional long-short term memory (Bi-LSTM) network for classification.

Main Results:

• CoMB-Deep achieved high accuracy in both binary (normal vs. abnormal) and multi-class (MB subclass) classification.
• Feature sets selected via CoMB-Deep's strategies significantly improved Bi-LSTM performance over individual deep features.
• Comparative analysis demonstrated CoMB-Deep's robustness and superior performance against related studies.

Conclusions:

• CoMB-Deep offers a reliable and automated solution for childhood medulloblastoma classification.
• The pipeline enhances diagnostic accuracy, reduces the risk of misdiagnosis, and accelerates the classification process.
• CoMB-Deep has the potential to assist pathologists, improve patient management, and lower healthcare costs.