A Novel Hybrid Approach for Classifying Osteosarcoma Using Deep Feature Extraction and Multilayer Perceptron

Area of Science:

• Computational pathology
• Artificial intelligence in oncology
• Digital histopathology

Background:

• Osteosarcoma is the most prevalent bone cancer, primarily affecting adolescents and young adults.
• Histopathology analysis of H&E-stained tissues presents challenges due to image complexity, noise, and class similarity.
• Accurate classification of osteosarcoma subtypes (nontumor, necrosis, viable tumor) is crucial for effective treatment planning.

Purpose of the Study:

• To develop and evaluate a hybrid deep learning framework for enhanced osteosarcoma tumor classification.
• To improve the diagnostic accuracy and efficiency of classifying osteosarcoma subtypes from whole slide images (WSIs).
• To integrate a robust feature selection mechanism for optimal model performance.

Main Methods:

• A hybrid framework combining pre-trained Convolutional Neural Network (CNN) models with a Multilayer Perceptron (MLP) classifier was developed.
• Transfer learning was employed using five CNN architectures as feature extractors on preprocessed WSIs.
• Recursive Feature Elimination (RFE) with a decision tree estimator was utilized for feature selection, followed by MLP classification with five-fold cross-validation.

Main Results:

• The proposed hybrid model achieved high accuracy: 95.2% for multiclass classification and 99.4% for binary classification.
• Feature selection analysis identified optimal criteria balancing execution time and classification accuracy.
• The model demonstrated superior performance compared to existing methods in osteosarcoma classification.

Conclusions:

• The developed hybrid AI model significantly improves the accuracy of osteosarcoma tumor classification from digital histopathology slides.
• This approach offers a valuable tool to assist clinicians in the diagnosis of osteosarcoma, potentially improving patient outcomes.
• The model's integration into a web application enables real-time predictions, facilitating clinical application.