From:  AI-integrated nanotherapeutics for schizophrenia: advancing precision oral drug delivery

 Reported diagnostic performance of selected AI-based approaches in SZ-related studies.

AI techniqueInput data & feature typeDiagnostic taskPerformanceReferenceAI technique
CNN on EEG spectrogram imagesMultichannel EEG → spectrogram conversionSZ vs. healthy control98% accuracy[34]CNN on EEG spectrogram images
CNN-LSTM on EEG signalsRaw EEG via CNN-LSTM architectureSZ diagnosis99.25% accuracy[33]CNN-LSTM on EEG signals
Lightweight 3D-CNN + ensemble on structural MRIT1-weighted 3D MRI volumesSZ vs. control92.22% accuracy[35]Lightweight 3D-CNN + ensemble on structural MRI
2D- & 3D-CNN on diffusion + structural MRIsMRI + DTI, pre-trained and scratch CNNsClassification~90–95% (AUC ~0.90)[36]2D- & 3D-CNN on diffusion + structural MRI
Return to article view

Note: the performance values summarized in this table are study-reported outcomes and should not be interpreted as directly comparable across modalities. The included studies differ in sample size, data type, diagnostic task, preprocessing method, model architecture, validation strategy, and outcome metric. Therefore, the table is intended to provide an overview of reported AI applications rather than a ranking of EEG-, MRI-, ERP-, or multimodal approaches. AI: artificial intelligence; AUC: area under the curve; CNN: convolutional neural network; DTI: diffusion tensor imaging; EEG: electroencephalography; ERP: event-related potential; LSTM: long short-term memory; MRI: magnetic resonance imaging; SZ: schizophrenia.