Premium
IC‐P1‐021: A novel web‐based system for automated computer classification in Alzheimer's dementia
Author(s) -
Duchesne Simon,
Crépeault Burt
Publication year - 2008
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.2008.05.032
Subject(s) - linear discriminant analysis , projection (relational algebra) , artificial intelligence , computer science , pattern recognition (psychology) , dementia , set (abstract data type) , machine learning , medicine , pathology , algorithm , disease , programming language
Background: Automated computer classification (ACC) techniques based on routine MRI provide aid to diagnostic information in an objective and reproducible fashion without being resource-intensive. To ensure simple yet reliable access, we propose a web-based ACC prototype tailored to the task of differentiating probable AD from normal aging. Methods: Input The system is based on anonymized T1weighted MR scans. Image quality control (QC) is performed manually on an 11-point scale. Processing The comparative ACC methodology exploits the differences in covariant patterns of Jacobian determinants and scaled grey-level intensity between subjects within a limited volume of interest centered on the medial temporal lobe. The classification result is obtained by comparing the resulting projection coordinates on relevant axes against a linear discriminant function. The nonlinear deformation field is also used to extract volumetric measures of specific temporal structures. Training data In the current instantiation the normative reference space for data projection was created using intensity and determinant features from 150 young controls. The linear discriminant function was obtained via forward stepwise linear discriminant analysis of a training set composed of 150 subjects: 75 patients with a diagnosis of probable AD and 75 age-matched controls without neurological or neuropsychological deficit. The AD subjects were individuals with mild to moderate probable AD recruited among outpatients seen at the IRCC Fatebenefratelli (Brescia, Italy). Longitudinal clinical assessment of all patients supported the diagnostic used as ground truth for the evaluation of the classification function. The final function is composed of 12 intensity and determinant features. Results: Reporting The report consists in three parts: a) QC information; b) quantitative information (e.g. Z-score of amygdalae and hippocampi with respect to retrospective data on age-matched CTRL); and c) summary classification information in support of clinical diagnosis. Conclusions: This web-based prototype interface is ready for routine clinical use for physicians working at the training site. Multi-site models will be made available in the next generation. The proposed technique is completely automated, requires no external expertise, and is completely objective. Acknowledgments: ICBM project, G. B. Frisoni (Brescia, Italy) for the data.