Schizophrenia (SZ) is a serious mental illness, yet only 10% of those affected find employment (Kahn, 2020). It shortens life expectancy by over 15 years, with suicide being a major cause (Peritogiannis et al., 2022). Risk factors for suicide include younger age, male gender, being single, living alone, unemployment, and high intelligence or education. Unrealized personal hopes and poor work performance further increase the risk. Antipsychotics effectively reduce psychosis, with two-thirds of patients achieving remission within 4 to 10 weeks and maintaining stability after a year (Kikkert et al., 2022). Diagnosis, based on clinical examination, can take up to two years due to the illness’s varied presentation. Early detection and treatment are essential to improve outcomes, reduce social costs, and ease family burdens.

Structural MRI (sMRI) is a powerful neuroimaging technique that allows researchers and clinicians to visualize the anatomy of the brain in great detail (Backhausen et al., 2022). Structural brain abnormalities in SZ are linked to clinical symptoms and underlying pathophysiology, aiding in early diagnosis, disease monitoring, and targeted treatments. sMRI plays a crucial role in understanding brain organization in SZ (Perlini et al., 2012). Patients with SZ often show reduced grey matter, particularly in the prefrontal cortex, temporal lobes, hippocampus, and thalamus (Chatterjee et al., 2020, Liu et al., 2020), regions associated with cognitive functions, emotion regulation, and memory, which are typically impaired in SZ. Longitudinal sMRI studies have shown that structural brain changes in SZ can progress over time (Yang et al., 2021, Merritt et al., 2021). This progression can be associated with the chronicity and severity of the illness. Structural abnormalities detected by sMRI often correlate with the severity of clinical symptoms (Kaur et al., 2020). Some studies have shown that antipsychotic treatment can lead to changes in brain volume, which may correlate with clinical improvement (Emsley et al., 2023).

Human brain activity and disorders can now be studied using the potent neuroimaging method known as functional MRI (fMRI). It is one of the main methods for demonstrating how different parts of the brain interact to carry out particular tasks. It offers a viable method for examining the interactions between geographically separated brain areas that are working on a task at the same time. The functional consistency of spatially distant regions is maintained through temporal coherence in the fMRI blood oxygenation level-dependent (BOLD) signal, which is not random (Salehi et al., 2024). Brain networks are made up of functionally connected regions. In neuroscience, the identification of functional activation in fMRI data during task or task-free “resting state” is crucial for pre-operative planning and the diagnosis of neuropsychiatric diseases such as SZ (Ahmad et al., 2023), autism (Giansanti, 2023), and Alzheimer’s disease (Sethuraman et al., 2023).

Group differences in sMRI reveal structural abnormalities in SZ, crucial for understanding its neuroanatomical basis and developing diagnostic biomarkers for early detection (Qi et al., 2022). fMRI highlights disruptions in functional connectivity, showing how SZ affects brain communication (Qiu et al., 2021). These functional differences can serve as biomarkers for diagnosing SZ, predicting its progression, and assessing treatment response. Comparing pre- and post-treatment sMRI and fMRI data helps evaluate how interventions like antipsychotics and therapy impact brain structure and function. Group differences in sMRI and fMRI can also reveal how SZ affects cognitive processes such as memory, attention, and executive function, thereby elucidating the cognitive deficits associated with the disorder. Furthermore, identifying correlations between brain structure and activity with specific symptoms can aid in developing symptom-specific treatments.

Explainable artificial intelligence (XAI) provides consumers with an explanation of the outcomes that artificial intelligence has anticipated (Dwivedi et al., 2023). This enables the identification of the primary influencing elements, comprehending the rationale behind the decision made in light of the machine learning (ML) model’s prediction result, and providing a clear and understandable explanation for the probable result. By addressing the “black box” issue, XAI enhances trust and efficiency, particularly in AI-driven systems like chatbots. Balancing user needs with analytical depth is key to improving AI communication. An XAI approach called Gradient-weighted Class Activation Mapping (Grad-CAM) (Song et al., 2023) is used to visualize the regions of an input signal that have the most influence on a given decision produced by a deep learning (DL) model. It creates a heatmap by utilizing the gradients of the goal concept and feeding it into the last convolutional layer to highlight significant areas in the input data. This enables industry professionals to comprehend and analyze sophisticated neural network models’ conclusions more effectively.

In this paper, we have correlated the results from ResNet-50 with Grad-CAM and group differences analysis of fMRI and sMRI. The significant structural variations for the classification of SZ from CN are identified by employing an XAI network for sMRI data. Significant group differences between SZ and CN groups are calculated for sMRI data. Additionally, group differences between SZ and CN during the N-back task related to fMRI data are examined. These structural and functional findings highlight relevant results, particularly the significance of the frontal lobe, which aids clinicians in the diagnosis of SZ.

The proposed paper’s major contributions are listed below.

The sMRI data is forwarded to the ResNet-50 and Grad-CAM an XAI to find significant regions of the brain for the classification of SZ from CN.

The group differences analysis between SZ and CN is calculated for sMRI and fMRI data to identify the structural abnormalities and the disruptions in functional activity in the brain region.

The results from both ResNet-50 with Grad-CAM and group differences analyses are correlated to explore the significance of the frontal lobe.

The performance of ResNet-50 and Grad-CAM is compared with various pre-trained models to explore the significance of the suggested architecture.

The proposed paper’s framework is as follows::The literature review related to SZ diagnosis is discussed in Section 2. Section 3 covers the proposed methodology associating ResNet-50 with Grad-CAM and group difference analysis for fMRI and sMRI. Section 4 illustrates the results and discussion. Finally, Section 5 concludes the paper.