Answers are grounded in this paper and cite their sources.
The EADC-ADNI Harmonized Protocol (HP) is a Delphi definition of manual hippocampal segmentation from magnetic resonance images (MRI). The project produced benchmark, standardized labels enabling training and qualification of future human tracers. However, training automated segmentation algorithms requires larger datasets representative of wider physiological variability. This work aims to produce a set of standardized labels allowing proper training of automated segmentation algorithms. We selected 135 ADNI images from different subjects, balanced by diagnosis, degree of medial temporal atrophy (Scheltens' visual scale, 1992), scanner type, magnet field strength, and age. We recruited tracers among HP segmentation experts based on excellent tracing performance as judged by volume ICC>0.94 or Jaccard>0.82 (Figure). We certified standardized labels as HP-compliant after centralized quality checks and corrections. Internal hippocampal CSF spaces were separately labelled, and excluded from hippocampal volume. We run t-tests and correlation analyses to evaluate volume differences across diagnoses and medial temporal atrophy. We have completed segmentations, double step quality check, corrections and final certification for N=68 subjects at 1.5T MRIs, with a single tracer. Uncorrected hippocampal volumes of patients differed significantly from controls: AD patients had 19% smaller volumes than controls (left: mm 3 2386, SD=510, p=0.001, right: 2427 SD=568, p=0.001); MCI had 14% smaller volumes than controls (left: 2576, SD=521, p=0.026, right: 2550 SD=614, p=0.013); Controls: left 2953, SD=590, right: 2994 SD=537. These raw volumes significantly correlated with atrophy severity at Scheltens' scale (Table).CSF spaces were larger in controls: AD, left: mm 3 16, SD=16, p=0.036, right: 31, SD=30, p=0.074; MCI, left: 17, SD=9, p=0.033, right: 29, SD=16, p=0.102; Controls, left, 34 SD=35, right: 34 SD=37. Two subjects (3%) presented with unusual anatomy. Example of compliance with the Harmonized Protocol, required to a new tracer to be a tracer for the Harmonized Benchmark Labels. In this case, the tracer had a Jaccard index of 0.87 with the mean contour of the benchmark labels provided by the Harmonized Protocol project.
No takes yet. Share an insight, caveat, or question.
Boccardi et al. (2013) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: