Premium
P1‐119: Prospective quality‐control monitoring in the context of a clinical trial
Author(s) -
Figurski Michal J.,
Waligórska Teresa,
Brylska Magdalena,
Korecka Magdalena,
Fields Leona,
Shah Nirali,
Pan Sarah,
Siemers Eric R.,
Lachno David R.,
Deckard Deanilee,
Dean Robert A.,
Trojanowski John Q.,
Shaw Leslie M.
Publication year - 2015
Publication title -
alzheimer's and dementia
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 6.713
H-Index - 118
eISSN - 1552-5279
pISSN - 1552-5260
DOI - 10.1016/j.jalz.2015.06.317
Subject(s) - context (archaeology) , medicine , replicate , standard operating procedure , biomarker , clinical trial , chemistry , operations management , mathematics , statistics , paleontology , biochemistry , economics , biology
enforce boundary smoothness. The technique’s validity was tested by comparing our results to the manual HarP reference segmentations on a sample of 100 subjects from the ADNeuroimaging Initiative (ADNI) database. We used a leave-one-out strategy to train and then test the segmentations. We used Dice similarity index for objects and correlation coefficients for volumes to verify compliance of automated to manual segmentations. Results: The results show that our method is bias-free with an average Dice similarity coefficient value of 0.803 and overall correlation coefficient of r1⁄40.95. Figure 1 compares the HC volumes computed from the segmentations made by anatomists (reference volumes) with the corresponding volumes computed from the automated segmentations (estimated segmentations). Two fits are shown, with and without intercept (fit parameters listed in Table 1). Both fits are virtually indistinguishable. Overlaid to a MRI image, the new contours often outperform those made by the anatomists in delineating the HC tissues in all three planes, due to the increased smoothness. Conclusions:Our automated segmentation algorithm was able to generate accurate HC volume measurements on a sample of the ADNI cohort that is heterogeneous in terms of ages, cognitive status, manufacturers, and atrophy levels. To our knowledge, this is the first study to assess the accuracy of an automated algorithm with the official release of HarP segmentation labels. Our results hold promise for the utilisation of automated segmentation in large clinical trials and in clinical practice where there is a growing need for biomarkers to support diagnosis and monitor progression of AD.