Supervised automatic procedure to identify new lesions in brain MR longitudinal studies of patients with multiple sclerosis

Abstract

Purpose: Identification of new enhancing lesions is a major endpoint of longitudinal brain magnetic resonance (MR) studies of multiple sclerosis (MS). To date, this is a visual, time-consuming procedure. We present here a supervised automated procedure (SAP) aimed at reducing the time needed to identify new MS enhancing lesions.

Materials and methods: The SAP uses an algorithm including Cartesian coordinates of the lesions to be compared, their area and a constant (k). The procedure was validated for enhancing lesions on T1-weighted spin-echo images after intravenous administration of 0.1 mmol/kg of paramagnetic contrast agent, randomly selected from a dataset of a longitudinal MR study on ten relapsing-remitting MS patients followed for 2-5 years. During the validation session, two readers decided by consensus whether two lesions, present on the same slice of two examinations performed on subsequent dates, were the same or not. In this way, k was calibrated to obtain the same result from both visual inspection and automatic algorithm output.

Results: After evaluating of 25+/-5 (mean+/-standard deviation) lesions in each of ten different sessions with correction of k value, the k value became a stable value (0.45+/-0.05).

Conclusions: Once the suitable value of k was found, SAP was able to identify new enhancing lesions, avoiding visual inspection, which is usually a lengthy procedure.