Machine Learning-Derived Pathomics Signature Predicts Prognosis in PCNSL

Overview

This study developed a novel pathomics score (Path-score) using machine learning on histological slides to predict prognosis in primary central nervous system lymphoma (PCNSL). The Path-score demonstrated strong correlation with treatment response and outperformed existing prognostic models when combined with clinical features in a nomogram.

Background

Primary central nervous system lymphoma (PCNSL) is a rare, aggressive extranodal non-Hodgkin lymphoma primarily involving the CNS. Despite advances, prognosis remains poor with 5-year overall survival rates between 22.3% and 35%. Current prognostic models like IELSG and MSKCC have limitations due to incomplete clinical data and inconsistent survival prediction. Computational pathology and machine learning offer new avenues to extract prognostic biomarkers from histological images, potentially improving outcome prediction in PCNSL.

Data Highlights

Key Findings

• A fully automated pipeline extracted quantitative pathomics features from H&E-stained slides of PCNSL patients.
• The Path-score derived via LASSO-Cox regression effectively stratified patients by survival outcomes.
• Path-score correlated significantly with initial treatment response, distinguishing responders from non-responders.
• A nomogram combining Path-score and clinical variables outperformed IELSG and MSKCC prognostic models in predicting overall survival.
• The study utilized two independent cohorts from Beijing Tiantan Hospital for model development and validation, ensuring robustness.

Clinical Implications

The Path-score offers a novel, objective biomarker derived from routine histopathology slides that can aid clinicians in prognostic stratification of PCNSL patients. Incorporating this score with clinical factors into a nomogram may improve individualized treatment planning and identify patients at higher risk of poor outcomes who may benefit from intensified therapy or closer monitoring.

Conclusion

This study demonstrates that machine learning-derived pathomics features from histological slides provide valuable prognostic information in PCNSL. The Path-score, especially when combined with clinical data, enhances survival prediction beyond existing models, representing a promising tool for clinical decision-making.

References

1. International Extranodal Lymphoma Study Group (IELSG) -- Prognostic Model for PCNSL
2. Memorial Sloan-Kettering Cancer Center (MSKCC) -- Prognostic Model for PCNSL
3. LASSO Regression in Survival Prediction -- Prior Studies
4. Digital Pathology and Machine Learning Applications in Oncology