Development and validation of an interpretable machine learning model for standard spleen volume prediction.

Area of Science:

• Medical Imaging and Diagnostics
• Machine Learning in Healthcare
• Hematology and Oncology

Background:

• Splenomegaly is a key indicator for various diseases, including hepatosplenomegaly and hematological disorders.
• Accurate assessment of splenic volume is crucial for diagnosis and treatment decisions.
• Individualized diagnosis requires a standard reference for splenic volume.

Purpose of the Study:

• To develop an interpretable machine learning (ML) model for evaluating standard splenic volume (SSV).
• To enhance personalized clinical decision-making for conditions involving splenomegaly.
• To provide a reliable reference for splenic volume assessment.

Main Methods:

• Retrospective analysis of 1,186 volunteers from a multicenter cohort.
• Evaluation of 11 ML algorithms, with SHapley Additive exPlanations (SHAP) for feature interpretation.
• Rigorous performance evaluation using RMSE, R², and comparison with existing formulas.

Main Results:

• A random forest (RF) model (ML_SSV) integrating 11 features showed high predictive accuracy (RMSE=22.6 mL, R²=0.80 in external validation).
• A simplified RF model (ML_SSVa) using 4 non-invasive parameters also demonstrated robust performance (RMSE=36.0 mL, R²=0.70).
• Both models outperformed existing formulas and are available as open-access web calculators.

Conclusions:

• Novel interpretable ML models for SSV assessment have been developed and validated.
• These models provide a reference baseline for individualized splenomegaly diagnosis.
• The tools enhance diagnostic accuracy and support data-driven clinical decisions.