Assessa is based on the LEAP (Learning Embeddings for Atlas Propagation) algorithm (Wolz et al 2010).
In this method, the brain of the subject being processed is compared to a reference database containing multiple images with pre-segmented brain regions. LEAP uses patented technology to select a subset of the reference images that are most similar to the subject image, specifically taking into account the degenerative changes observed in the demented brain. Aligning all selected reference images to the subject image allows the algorithm to obtain an estimate of the hippocampus localisation in the subject image.
LEAP has been shown to be able to predict time to dementia in subjects with MCI with an accuracy better than that of radiological scoring of Medial Temporal Atrophy (van Rossum et al 2012, Clerx et al 2013, Hill et al 2013 and to be able to add complementary information to an analysis of cerebrospinal fluid, CSF (Vos et al 2012). LEAP can be applied to multiple brain structures including the hippocampus, amygdala, parahippocampal gyrus and lateral ventricles (Wolz 2011), and a combined analysis has been shown to be able to increase the predictive power over that of hippocampal volume alone (Wolz et al 2013).
LEAP was one of the algorithms included in the EMA qualification of Low Hippocampal Volume as a biomarker to enrich clinical trials of AD in the pre-dementia phase (EMA 2011, Hill et al 2013). The test:re-test performance of the LEAP algorithm has been characterized both on repeat scans on a single scanner, and moving the subject between 1.5T and 3T scanners, showing the robustness of the method towards the typical variability in acquisition protocols found between different clinical centers (Wolz et al 2013b).