Table 2.
Neuroimaging-based brain age studies for psychiatric disorders.
| First Author [ref.] |
Year | Cohort | Imaging Modality | ML Algorithm | Main Findings |
|---|---|---|---|---|---|
| Schizophrenia and Psychosis | |||||
| Koutsouleris [57] | 2014 | 141 SZ, 104 MDD, 57B PD, 89 ARMS, 127 HCs | T1WI | SVR | BAG: +5.5 yr in SZ, +4.0 yr in MDD, +3.1 yr in BPD, +1.7 yr in ARMS. |
| Schnack [58] | 2016 | 341 SZ, 386 HCs | T1WI | SVR | BAG: +3.36 yr in SZ, acceleration just after illness onset |
| Nenadic [59] | 2017 | 45 SZ, 22 BPAD, 70 HCs | T1WI | RVR | BAG: +2.56 yr in SZ, no significance in BPAD |
| Kolenic [60] | 2018 | 120 FEP, 114 HCs | T1WI | RVR | BAG: +2.64 yr in FES, associated with obesity |
| Hajek [62] | 2019 | 43 FES, 43 HCs, 96 offspring of BPAD (48 affected, 48 unaffected), 60 HCs | T1WI | RVR | BAG: +2.64 yr in FES, no significance in early BPAD |
| Chung [61] | 2019 | 476 CHR | N/A | N/A | BAG predicts conversion to psychosis in a univariate analysis but not in a multivariate analysis |
| Shahab [63] | 2019 | 81 SZ, 53 BPAD, 91 HCs | T1WI, DTI | RF | BAG: +7.8–8.2 yr in SZ, no significance in BPAD |
| Kuo [64] | 2020 | 26 SZ, 30 MDD, 19AD, 109 HCs | T1WI | LASSO, ICA | BAG: +5.69 yr in SCZ, +3.25 yr in AD, no significance in MDD. Association with large-scale structural covariance network |
| Tønnesen [65] | 2020 | 668 SZ, 185 BPAD, 990 HCs | DTI | XGBoost | Increased BAG in SZ (Cohen’s d = −0.29) and BPAD (Cohen’s d = 0.18) |
| Lee [66] | 2021 | 90 SZ, 200 HCs, 76 SZ, 87 HCs | T1WI | OLS, Ridge, LASSO, Elastic-Net, SVR, RVR | BAG: +3.8–5.2yr in SZ cohort 1, +4.5–11.7 yr in SZ cohort 2. Algorithm choice can be a cause of inter-study variability. |
| Lieslehto [67] | 2021 | 29 SZ, 61 HCs | T1WI | SVR | BAG: +1.3 yr at baseline, +7.7 yr at follow-up in SZ. It was suggested that BA captured treatment-related and global brain alterations. |
| McWhinney [68] | 2021 | 183FEP, 155 HCs | T1WI | RVR | BAG: +3.39 yr in FEP at baseline, longitudinal worsening was associated with clinical outcomes or higher baseline BMI |
| Teeuw [69] | 2021 | 193 SZ, 218 HCs | T1WI | SVR | BAG: correlation with polygenic risk, no correlation with epigenetic aging |
| Wang [70] | 2021 | 166 SZ, 107 HCs | DTI | RF | BAG: +5.903 in SZ >30 yrs old. Association with working memory and processing speed |
| Xi [71] | 2021 | 60 FES, 60 HCs | DTI | RVR | BAG: +4.932 yr in FES, +2.718. Decreased BAG after early medication |
| Demro [72] | 2022 | 163 psychosis, 103 relatives, 66 HCs | T1WI | SVR/RF | BAG increase in psychosis more than HCs or relatives. Associated with cognition or schizotypal symptoms in relatives |
| Mood disorders | |||||
| Bestteher [73] | 2019 | 38 MDD, 40 HCs | T1WI | RVR | BAG: no significant change in MDD |
| Van Gestel [74] | 2019 | 84 BPAD, 45 HCs | T1WI | RVR | BAG: +4.28 yr in BPAD without Li treatment, no significance in BPAD with Li treatment or HCs |
| de Nooij [75] | 2019 | 283AYA | T1WI | RVR | Reduction of BAG in young high-risk individuals who developed a mood disorder over 2-yr follow-up |
| Christman [76] | 2020 | 76 MDD (middle-age), 118 MDD (elderly), 130 HCs | T1WI | CNN | BAG: +3.69 yrs in geriatric MDD, no increase in mid-life MDD. Associated with cognitive and functional deficits in elderly |
| Ahmed [77] | 2021 | 95 late-life depression | T1WI | CNN | BAG: +4.36 yrs in late-life depression. Not associated with treatment response. |
| Ballester [78] | 2021 | 160 MDD, 111 HCs | T1WI | GPR | BAG: higher in older MDD than in younger MDD, associated with BMI in MDD, not associated with treatment response |
| Han [79] | 2021 | 2675 MDD, 4314 HCs | T1WI | Ridge regression | BAG: +1.08 yr in MDD with no specific association with clinical characteristics |
| Han [80] | 2021 | 220 MDD/Anxiety, 65 HCs | T1WI | Ridge regression | BAG: +2.78 yr in MDD, +2.91 yr in Anxiety. Association with somatic symptoms (+4.21 yr) and antidepressant use (−2.53 yr) |
| Dunlop [81] | 2021 | 109 MDD, 710 HCs | fMRI | SVR | BAG: +2.11 yr in MDD, associated with impulsivity and symptom severity |
| Others | |||||
| Liu [82] | 2022 | 90 OCD, 106 HCs | T1WI | GPR | BAP: +0.826 yr in OCD, associated with disease duration |
| Niu [83] | 2022 | 70 SP, 77 SAD, 70 MDD, 44 PTSD, 48 ODD, 81 ADHD | T1WI | Ridge regression | Multidimensional brain-age index is sensitive to distinct regional change patterns |
| Ryan [84] | 2022 | 1618 SMI, 11,849 HCs | DTI | RF, gradient boosting regression, LASSO | Additive effect of SMI and cardiometabolic disorders on brain aging, the greater effect of SMI than CMD |
| Comprehensive | |||||
| Kaufmann [85] | 2019 | 10,141 patients, 35,474 HCs | T1WI | XGBoost | BAG: d = +1.03 in dementia, +0.41 in MCI, +0.10 in MDD, +0.74 in MS, +0.29 in BPAD, +0.51 in SZ, +0.06 in ADHD, +0.07 in ASD |
| Bashyam [86] | 2020 | 353 AD, 833 MCI, 387 SZ, 12,689 HCs | T1WI | CNN | Successful discrimination for neuropsychiatric disorders |
| Kolbeinsson [87] | 2020 | 12,196 people who had not been stratified for health | T1WI | CNN | Identified risk factors, e.g., MS, diabetes, and beneficial factors, e.g., physical strength |
| Rokicki [88] | 2021 | 54 AD, 90 MCI, 56 SCI, 159 SZ, 135 BPAD, 750 HCs | T1WI, T2WI, ASL | RF | Highest accuracy by multimodal imaging model |
AD: Alzheimer’s disease, ARMS: at-risk mental state, ASL: arterial spin labeling, AYA: adolescence and young adult, BAG: brain age gap, BPAD: bipolar affective disorder, BPD: borderline personality disorder, CHR: clinical high-risk state for psychosis, CMD; cardiometabolic disease, CNN: convolutional neural network, DTI: diffusion tensor imaging, FEP: first episode psychosis, FES: first-episode schizophrenia, fMRI: functional MRI, GPR: Gaussian process regression, HCs: healthy controls, ICA: independent component analysis, MDD: major depressive disorder, ML: machine learning, OCD: obsessive-compulsive disorder, ODD: oppositional defiant disorder, OLS: ordinary least squares, PTSD: posttraumatic stress disorder, RF: random forest, RVR: relevance vector regression, SAD: social anxiety disorder, SMI: severe mental illness, SP: specific phobias, SVR: support vector regression, SZ: schizophrenia, T1WI: T1-weighted image, T2WI: T2-weighted image.