Preliminary assessment of an MRI-based grading system for leptomeningeal disease: an exploratory prognostic framework

Background: Current diagnostic approaches of leptomeningeal disease (LMD) rely heavily on cerebrospinal fluid (CSF) cytology, which shows significant limitations and the requirement for invasive procedures. We aim to develop an MRI-based grading scores for LMD diagnosis and prognosis that address current diagnostic limitations and provide standardized, reproducible assessment criteria. Methods: We conducted a retrospective analysis of 32 adult cancer patients evaluated for suspected LMD. Two experienced neuroradiologists independently assessed MRI studies using our novel grading system, which incorporates leptomeningeal enhancement/intensity patterns (grades 1–6), Evans index for hydrocephalus assessment, brain metastases characteristics, and spinal involvement. Confirmation of LMD cases was employed using dual confirmation approach combining CSF cytology and follow-up MRI. Results: Our MRI grading system demonstrated promising inter-observer performance. Inter-rater reliability between two attending level neuroradiologists was excellent (ICC = 0.953, P-value < 0.001) using a cutoff score of 2 or higher, the system demonstrated comparable performance. Risk stratification analysis revealed clear prognostic value, with mortality rates of 8.6% for low-risk patients (Grade 1–2), 50% for medium-risk patients (Grade 3–4), and 80.0% for high-risk patients (Grade 5 +). The Kaplan–Meier survival curves demonstrate a statistically significant difference in overall survival between patients with varying grades (p-value of 0.00011). Notably, survival probability drops steeply in the Grade 5 + group early on, suggesting that higher LMD burden is associated with rapid clinical deterioration. In contrast, low risk patients appear to have a more indolent course. Conclusions: Our preliminary findings detail a promising approach in evaluating LMD patients which offers valuable prognostic information for clinical decision making. Furthermore, the high inter-rater reliability across various tumor types further encourages the potential utility of this approach, although further research on a broader population is needed before clinical implementation.