Brain imaging and behavior · 2026
From voxels to models: a multi-center voxel-based morphometry study identifying robust biomarkers for deep learning classification of schizophrenia.
СтатьяМногоцентровое исследованиеНазвание и аннотация: Перевод пока не подготовлен
Обработка статьи · этапы, ошибки и управление
Загрузка состояния…
Ошибки и ограничения
История · было → стало
Состояние обновляется каждые 5 секунд, пока блок открыт. Процент показывается только там, где известен объём работы. Пауза останавливает продолжение этапа; текущий запрос к источнику может завершиться.
Adamu Mohammed Jajere, Qiang Li, Nyatega Charles Okanda
Аннотация
BACKGROUND: The application of deep learning in psychiatric disorders such as schizophrenia has lagged behind neuro-oncology, primarily due to a lack of identifiable neuroanatomical targets and concerns regarding generalizability across sites.
METHODS: To address this, we analyzed structural MRI data from a multi-site study of 401 subjects (194 schizophrenia, 207 healthy) across three independent datasets (COBRE, FBIRN, SLIM). We employed a two-stage conservative voxel-based morphometry (VBM) approach where neuroanatomical regions identified in a discovery dataset (COBRE) were rigorously validated in independent replication datasets (FBIRN, SLIM). The neuroanatomical features that survived this validation were then used to train a support vector machine (SVM) classifier with strictly nested cross-validation to prevent data leakage.
RESULTS: Fifteen robust structural biomarkers were identified, consistently showing gray matter reductions in frontal and temporal regions, white matter alterations, and ventricular enlargement across all datasets. The SVM classifier achieved an area under the curve (AUC) of 0.94 with 89.5% accuracy in initial analysis, with strictly nested cross-validation confirming robust performance (AUC = 0.93, accuracy = 88.7%). The model maintained high performance in leave-one-dataset-out validation (AUC = 0.90-0.93), demonstrating generalizability across unseen data from differing sites. A label-shuffling permutation test confirmed that the feature set reflects genuine neuroanatomical signal rather than statistical artifacts (p < 0.001).
CONCLUSION: This study demonstrates a robust, multi-site validated set of structural biomarkers associated with schizophrenia, enabling highly accurate case-control classification. The identified feature set provides a candidate neurobiological benchmark to inform and validate future deep learning models for automated diagnostic support in psychiatry, while acknowledging the need for future multi-diagnostic validation to establish differential diagnostic utility.
METHODS: To address this, we analyzed structural MRI data from a multi-site study of 401 subjects (194 schizophrenia, 207 healthy) across three independent datasets (COBRE, FBIRN, SLIM). We employed a two-stage conservative voxel-based morphometry (VBM) approach where neuroanatomical regions identified in a discovery dataset (COBRE) were rigorously validated in independent replication datasets (FBIRN, SLIM). The neuroanatomical features that survived this validation were then used to train a support vector machine (SVM) classifier with strictly nested cross-validation to prevent data leakage.
RESULTS: Fifteen robust structural biomarkers were identified, consistently showing gray matter reductions in frontal and temporal regions, white matter alterations, and ventricular enlargement across all datasets. The SVM classifier achieved an area under the curve (AUC) of 0.94 with 89.5% accuracy in initial analysis, with strictly nested cross-validation confirming robust performance (AUC = 0.93, accuracy = 88.7%). The model maintained high performance in leave-one-dataset-out validation (AUC = 0.90-0.93), demonstrating generalizability across unseen data from differing sites. A label-shuffling permutation test confirmed that the feature set reflects genuine neuroanatomical signal rather than statistical artifacts (p < 0.001).
CONCLUSION: This study demonstrates a robust, multi-site validated set of structural biomarkers associated with schizophrenia, enabling highly accurate case-control classification. The identified feature set provides a candidate neurobiological benchmark to inform and validate future deep learning models for automated diagnostic support in psychiatry, while acknowledging the need for future multi-diagnostic validation to establish differential diagnostic utility.