Unveiling the invisible: How cutting-edge neuroimaging transforms adolescent depression diagnosis

Table 1 Results of previous studies using neuroimaging techniques, machine learning, and deep learning in the context of major depressive disorder

Neuroimaging technique	Machine learning approaches	Deep learning approaches	Findings/applications	Ref.
fMRI	SVM, XGBoost	CNN	Identification of functional connectivity alterations, prediction of treatment response, generalizable markers across imaging sites	Anderson et al[21], Shi et al[22], Yamashita et al[23]
sMRI	PCA, ICA	Autoencoders	Detection of gray matter volume reductions, feature extraction, and identification of structural biomarkers	Zhang et al[24], Ho et al[25]
PET	Feature selection		Evaluation of altered glucose metabolism, identification of treatment-specific biomarkers	McGrath et al[26], Kang and Cho[27]
EEG	Time-series analysis	CNN, RNN, LSTM	Analysis of altered neural oscillations, connectivity patterns, and prediction models using EEG-based approaches	Uyulan et al[28], Yang et al[29]

fMRI: Functional magnetic resonance imaging; SVM: Support vector machines; CNN: Convolutional neural networks; PCA: Principal component analysis; ICA: Independent component analysis; PET: Positron emission tomography; EEG: Electroencephalography; RNN: Recurrent neural network; LSTM: Long short-term memory model.

Table 2 Demographic and clinical characteristics

Variable	MDD group (n = 44)	Control group (n = 43)	P value
Age (years)	15 ± 2	15 ± 2	> 0.05
Sex (male/female)	10/34	11/32	> 0.05
Education (years)	10 ± 3	10 ± 3	> 0.05
HAMD-17 Score	21 ± 5	5 ± 3	< 0.05

MDD: Major depressive disorder; HAMD-17: Hamilton Depression Rating Scale-17.