Using a ResNet-18 Network to Detect Features of Alzheimer’s Disease on Functional Magnetic Resonance Imaging: A Failed Replication. Comment on Odusami et al. Analysis of Features of Alzheimer’s Disease: Detection of Early Stage from Functional Brain Changes in Magnetic Resonance Images Using a Finetuned ResNet18 Network. Diagnostics 2021, 11, 1071

• Published in: Diagnostics (Basel) Vol. 12; no. 5; p. 1094
• Main Authors: Nicholas, Peter J., To, Alex, Tanglay, Onur, Young, Isabella M., Sughrue, Michael E., Doyen, Stéphane
• Format: Journal Article
• Language: English
• Published: Switzerland MDPI AG 27.04.2022; MDPI
• Subjects: Accuracy; Alzheimers disease; Classification; Cognition & reasoning; Cognitive ability; Comment; Datasets; Deep learning; Experiments; Magnetic resonance imaging; Methods; n/a; Patients
• ISSN: 2075-4418; 2075-4418
• DOI: 10.3390/diagnostics12051094

[4] utilizing deep learning based-methods to predict MCI from AD and controls based on resting-state functional magnetic resonance imaging (rsfMRI) with near 100% accuracy, outperforming any previously reported prediction model. According to baseline diagnosis, 436 subjects were controls, 199 had Mild Cognitive Impairment (MCI), 88 had early MCI (EMCI), 37 had late MCI (LMCI), and 62 had Alzheimer’s Disease (AD). [...]the train and test sets could have slices from the same subject. Nonetheless, the stark contrast in results between the two hold-out strategies demonstrates the importance of using test sets which have never been encountered by the model and avoiding data leakage, which may ultimately compromise the applicability of the results.