Schizophrenia (SZ) is a severe form of neurological disorder impairing a subject’s functioning on social, emotional, and cognitive levels. Despite the urgency for early detection and treatment for better treatment prospects, conventional approaches in diagnosis most of the time are subjective. On the other hand, fMRI that can take brain activity is fast becoming one of the greatest research tools about SZ. In any case, to date, automated methods remain in dire need concerning diagnosing SZ patients through fMRI data. This work provides a de novo AI framework that uses axial fMRI data from the human brain to detect SZ. It combines ant colony optimization for hyperparameter tweaking with differential convolutional neural networks, which enhances the classification accuracy and the capacity of the model to learn complex features. To grasp different patterns of brain activity, the DCNN architecture employs multikernel convolutional layers along with residual connections, and ACO optimizes crucial hyperparameters. Our approach, which used fivefold cross-validation, showed great precision, recall, and F1 scores with an average accuracy of 98.43%. These findings indicate a potential direction for further clinical applications that our proposed framework may serve as a useful tool for the early diagnosis of schizophrenia.