Machine Learning-Based Classification of Neuroblastoma Risk Groups and MYCN Status

Area of Science:

• Computational pathology
• Digital pathology
• Artificial intelligence in oncology

Background:

• Artificial intelligence (AI) shows promise in analyzing histological features from whole slide images (WSIs) for tumor risk classification.
• Integrating histology with molecular data can improve risk stratification and clinical management for pediatric cancers like neuroblastoma.
• This study evaluates a deep learning AI model for classifying neuroblastoma patients by risk group and MYCN amplification status.

Purpose of the Study:

• To assess the performance of a deep learning AI model in classifying neuroblastoma patients into distinct risk groups using WSIs.
• To determine the AI model's capability in predicting MYCN amplification status from WSIs.
• To explore AI's potential as a complementary tool for neuroblastoma risk stratification.

Main Methods:

• Digital scanning of hematoxylin and eosin-stained slides from neuroblastoma cases across various risk groups and MYCN amplification statuses.
• Processing of WSIs using deep learning-based segmentation and classification algorithms.
• Development of two patch-level datasets for predicting MYCN amplification and classifying patients into four risk groups.

Main Results:

• The deep learning model achieved 98% accuracy in classifying neuroblastoma patient risk groups on slide-level tasks.
• The model demonstrated 99% accuracy in distinguishing MYCN amplified from non-amplified tumors.
• Within high-risk neuroblastoma cases, the model achieved 96% accuracy for MYCN amplification status prediction.

Conclusions:

• AI models can accurately classify neuroblastoma risk and predict MYCN amplification status from WSIs.
• These AI tools offer a promising complementary approach for pathologists in risk stratification.
• AI can be particularly valuable when histological examination is limited or molecular testing for MYCN amplification is inaccessible.