A Dual-Feature Framework for Enhanced Diagnosis of Myeloproliferative Neoplasm Subtypes Using Artificial Intelligence

Area of Science:

• Hematology
• Computational Pathology
• Machine Learning

Background:

• Philadelphia chromosome-negative (Ph-negative) myeloproliferative neoplasms (MPNs) like essential thrombocythemia, polycythemia vera, and primary myelofibrosis pose diagnostic challenges due to overlapping features.
• Current diagnostic methods (imaging, histopathology) suffer from interobserver variability and subjectivity, leading to delayed diagnoses.

Purpose of the Study:

• To develop and validate a novel computational framework for improved classification of Ph-negative MPNs.
• To integrate handcrafted and deep learning-based features for enhanced diagnostic accuracy.

Main Methods:

• A framework combining handcrafted (morphological, textural) and automatic (deep learning) feature extraction from histopathological images was developed.
• Machine learning models were trained using these features, with hyperparameter optimization via Optuna.
• Performance was evaluated using precision, recall, F1 score, accuracy, specificity, and weighted average.

Main Results:

• The integrated framework achieved a high mean weighted average of 0.9969, outperforming individual handcrafted (0.9765) and deep learning features (0.9686).
• Statistical analysis confirmed the robustness and reliability of the classification results.
• The combined approach demonstrated superior performance in classifying Ph-negative MPNs.

Conclusions:

• Combining domain-specific knowledge (handcrafted features) with data-driven approaches (deep learning) significantly enhances diagnostic accuracy for Ph-negative MPNs.
• This integrated framework offers a promising tool to support clinical decision-making and improve patient outcomes.
• Further research is needed to address challenges like feature distribution assumptions.