We built and validated a deep learning algorithm predicting the individual diagnosis of Alzheimer's disease (AD) and mild cognitive impairment who will convert to AD (c-MCI) based on a single cross-sectional brain structural MRI scan. Convolutional neural networks (CNNs) were applied on 3D T1-weighted images from ADNI and subjects recruited at our Institute (407 healthy controls [HC], 418 AD, 280 c-MCI, 533 stable MCI [s-MCI]). CNN performance was tested in distinguishing AD, c-MCI and s-MCI. High levels of accuracy were achieved in all the classifications, with the highest rates achieved in the AD vs HC classification tests using both the ADNI dataset only (99%) and the combined ADNI + non-ADNI dataset (98%). CNNs discriminated c-MCI from s-MCI patients with an accuracy up to 75% and no difference between ADNI and non-ADNI images. CNNs provide a powerful tool for the automatic individual patient diagnosis along the AD continuum. Our method performed well without any prior feature engineering and regardless the variability of imaging protocols and scanners, demonstrating that it is exploitable by not-trained operators and likely to be generalizable to unseen patient data. CNNs may accelerate the adoption of structural MRI in routine practice to help assessment and management of patients.

Cognitive impairment affects 40% to 65% of patients with multiple sclerosis (MS) and contributes significantly to their disability status.^1,2 Although it is more frequent and severe in the progressive forms of MS, such an impairment can be present since the early clinical stages of the disease^1,2 and typically includes deficits of long-term memory, attention and concentration, executive functioning, efficiency of information processing, and processing speed.^1-4

Objective: To assess cortical gray matter (GM) changes in MS and establish their relevance to clinical disability and to inflammatory changes of white matter (WM) in patients with the relapsing–remitting (RR) and primary progressive (PP) forms of the disease. Methods: Conventional MRI examinations were obtained in patients with definite MS who had either the RR or the PP form of the disease. An automated analysis tool was used with conventional T1-weighted MR images to obtain total and cortical brain volumes normalized for head size. Total brain lesion load was estimated on conventional proton density and T2-weighted MR images. The relationship between volumetric MR measures and scores of clinical disability was assessed. Results: Normalized cortical volumes (NCV) were lower for both RR and PP MS patients than for normal control subjects (p < 0.001) but were similar between the two patient groups (p > 0.5). NCV decreases in both patients groups were detected even in those patients with short disease duration (<5 years; p < 0.001 in RR MS and p < 0.05 in PP MS) and minimal brain lesion volume (<5 mL; p < 0.0001 in RR MS and p < 0.005 in PP MS). Measures of NCV in individual patients were negatively correlated with T2-weighted lesion volume (r = −0.47, p < 0.001) and disease duration (r = −0.25, p < 0.05) only in the patients with RR MS. NCV correlated with Expanded Disability Status Scale scores across all of the patients, but the strength of the correlation was stronger (p < 0.05) for PP (r = −0.64, p < 0.0001) than for RR (r = −0.27, p = 0.04) MS patients. Conclusions: These data confirm substantial neocortical volume loss in MS patients and suggest that neocortical GM pathology may occur early in the course of the disease in both RR and PP MS patients and contribute significantly to neurologic impairment. Although a proportion of this neocortical pathology may be secondary to WM inflammation, the extent of the changes suggests that, especially in patients with PP MS, an independent neurodegenerative process also is active.

Objective To determine whether brain atrophy and lesion volumes predict subsequent 10 year clinical evolution in multiple sclerosis (MS). Design From eight MAGNIMS (MAGNetic resonance Imaging in MS) centres, we retrospectively included 261 MS patients with MR imaging at baseline and after 1–2 years, and Expanded Disability Status Scale (EDSS) scoring at baseline and after 10 years. Annualised whole brain atrophy, central brain atrophy rates and T2 lesion volumes were calculated. Patients were categorised by baseline diagnosis as primary progressive MS (n=77), clinically isolated syndromes (n=18), relapsing–remitting MS (n=97) and secondary progressive MS (n=69). Relapse onset patients were classified as minimally impaired (EDSS=0–3.5, n=111) or moderately impaired (EDSS=4–6, n=55) according to their baseline disability (and regardless of disease type). Linear regression models tested whether whole brain and central atrophy, lesion volumes at baseline, follow-up and lesion volume change predicted 10 year EDSS and MS Severity Scale scores. Results In the whole patient group, whole brain and central atrophy predicted EDSS at 10 years, corrected for imaging protocol, baseline EDSS and disease modifying treatment. The combined model with central atrophy and lesion volume change as MRI predictors predicted 10 year EDSS with R²=0.74 in the whole group and R²=0.72 in the relapse onset group. In subgroups, central atrophy was predictive in the minimally impaired relapse onset patients (R²=0.68), lesion volumes in moderately impaired relapse onset patients (R²=0.21) and whole brain atrophy in primary progressive MS (R²=0.34). Conclusions This large multicentre study points to the complementary predictive value of atrophy and lesion volumes for predicting long term disability in MS.