Predicting Prognosis of Early-Stage Mycosis Fungoides with Utilization of Machine Learning

Abstract

Mycosis fungoides (MF) is the most prevalent type of cutaneous T cell lymphomas. Studies on the prognosis of MF are limited, and no research exists on the potential of artificial intelligence to predict MF prognosis. This study aimed to compare the predictive capabilities of various machine learning (ML) algorithms in predicting progression, treatment response, and relapse and to assess their predictive power against that of the Cox proportional hazards (CPH) model in patients with early-stage MF. The data of patients aged 18 years and over who were diagnosed with early-stage MF at Ankara University Faculty of Medicine Hospital from 2006 to 2024 were retrospectively reviewed. ML algorithms were utilized to predict complete response, relapse, and disease progression using patient data. Of the 185 patients, 94 (50.8%) were female, and 91 (49.2%) were male. Complete response was observed in 114 patients (61.6%), while relapse and progression occurred in 69 (37.3%) and 54 (29.2%) patients, respectively. For predicting progression, the Support Vector Machine (SVM) algorithm demonstrated the highest success rate, with an accuracy of 75%, outperforming the CPH model (C-index: 0.652 for SVM vs. 0.501 for CPH). The most successful model for predicting complete response was the Ensemble model, with an accuracy of 68.89%, surpassing the CPH model (C-index: 0.662 for the Ensemble model vs. 0.543 for CPH). For predicting relapse, the decision tree classifier showed the highest performance, with an accuracy of 78.17%, outperforming the CPH model (C-index: 0.782 for the decision tree classifier vs. 0.505 for CPH). The results suggest that ML algorithms may be useful in predicting prognosis in early-stage MF patients.