Integrated Multi-Criteria Decision Analysis for Enhanced ‎Lymphoma Diagnosis

Abstract

Lymphoma diagnosis remains a clinical challenge due to its biological heterogeneity, concurrent symptoms, and often uncertain initial manifestations. This study proposes an integrated data-driven system combining supervised machine learning and Multi-Criteria Decision Analysis (MCDA) to enhance clinical interpretability and diagnostic accuracy. The method ranks patient cases based on the relative severity of diagnostic factors such as CRP, LDH, hemoglobin, WBC, platelet count, tumor size, and age using an entropy-weighted TOPSIS model. Two thousand simulated patient profiles were created in a clinically informative dataset for validating and testing the system. The same set of features was used for training a Random Forest classifier to gauge the decision model's robustness. The classifier obtained an AUC of 0.53 and a prediction accuracy of 54.75%. Performance metrics were highly aligned with MCDA-generated rankings , and the three most important diagnostic markers – LDH, CRP, and hemoglobin – were always among both the machine learning models and the MCDA rankings. Complementary approaches like Principal Component Analysis (PCA), correlation heatmaps, ROC plots, and decision tree visualizations supported the models' structure and interpretability. The results show that the combination of MCDA and data-driven classification has the promise to aid the creation of transparent, flexible, and clinically meaningful diagnostic systems. This hybrid system supports future precision medicine uses, such as integration with imaging modalities, longitudinal monitoring of patients, and molecular data.