We enrolled 150 participants, with 50 in each group: Real-SCZ, Sham-SCZ, and healthy control (HC). All patients were at their first episode of schizophrenia and were diagnosed according to DSM-5 criteria (32). The two schizophrenia groups were matched for baseline clinical indicators and symptoms, with no significant differences between them. All patients were within two weeks of their initial exposure to antipsychotic treatment, with antipsychotic dosages converted to olanzapine equivalents using the Defined Daily Dose (DDD) system (33).

The HC group was recruited through community advertisements and screened to exclude any history of psychiatric or neurological disorders. All participants were right-handed, had normal or corrected-to-normal vision, and were free from concurrent neurological or severe systemic illnesses. Exclusion criteria for all participants included a history of traumatic brain injury, substance use disorders, or contraindications to MRI. Both groups excluded individuals with an intelligence quotient (IQ) below 70, estimated by the Wechsler Abbreviated Scale of Intelligence (34). The study was approved by the local Research Ethics Committee and conducted in accordance with the Declaration of Helsinki. Written informed consent was obtained from all participants prior to the commencement of the study. The study was conducted from January 2023 to March 2024. This period included all phases of the study: participant recruitment, data collection, and the subsequent analysis of behavioral and neuroimaging data. The detailed experimental design can be seen in Figure 1 .

The dual-choice oddball task requires participants to respond to two types of stimuli. One stimulus appears with a higher probability, approximately 85%, which is represented by the letter “W” in this experiment. The other stimulus appears with a lower probability, around 15%, and is represented by the letter “M” in this experiment. These stimuli are presented in random order, making the low-probability stimulus appear unexpectedly to the participant. The task requires participants to press the left key on the keyboard when they see the letter “W” and the right key when they see the letter “M.”

Participants received rTMS modulation over the left dorsolateral prefrontal cortex (DLPFC) according to the following protocol. Each rTMS session involved the delivery of 30 trains of TMS pulses at 10 Hz for 10 seconds (100 pulses per train) with a 20-second inter-train interval, resulting in a total of 3000 pulses per session for a total duration of 15 minutes. The stimulation intensity was set at 80% of the resting motor threshold, which was determined as the minimal intensity of stimulation that evoked an electromyographic response of 550 mV in the first dorsal interosseous muscle of the hand contralateral to the stimulated hemisphere in at least 5 out of 10 trials. For individualized stimulation (35), the MNI coordinates of the left DLPFC (approximately (–26, 38, 44), based on prior studies targeting this region) were adjusted for each participant using the inverse normalization method embedded in SPM12 (i.e., normalization based on the inverse deformation field). All rTMS sessions were guided by a neuronavigation system to ensure precise targeting. Once the location of the stimulation target was determined, the coordinates were entered into the neuronavigation system, which provided continuous guidance throughout the sessions, ensuring precise and consistent delivery of stimulation to the designated cortical area. For the Sham group, the rTMS coil was positioned at the same MNI coordinates of the left DLPFC as in the Real group, and the neuronavigation system was used to guide the coil placement. However, the coil was angled at 90 degrees away from the scalp, preventing direct brain stimulation while still producing auditory clicks and scalp sensations similar to the Real rTMS treatment. This setup ensured that participants in the Sham group experienced sensory feedback comparable to the Real group, such as the sound of the TMS pulses and the mild scalp sensations, without receiving active stimulation to the brain. This approach was used to maintain blinding of the participants and minimize placebo effects.

Functional MRI data were collected during the oddball task using a Philips 3-Tesla MRI scanner with a 32-channel head coil. BOLD (Blood Oxygen Level Dependent) signals were acquired using an echo-planar imaging (EPI) sequence with the following parameters: repetition time (TR) = 2000 ms, echo time (TE) = 30 ms, flip angle = 90°, field of view (FOV) = 220 mm × 220 mm, slice thickness = 3 mm, and 45 axial slices covering the entire brain. Each fMRI session included both task and rest periods. Preprocessing of fMRI data was conducted using SPM12 (Statistical Parametric Mapping). The preprocessing steps included discarding the first 10 time points, slice timing correction, realignment, normalization to the MNI template, and spatial smoothing with a 6 mm full-width at half-maximum (FWHM) Gaussian kernel.

The fMRI data were analyzed using a mixed-effects model in a two-stage process. In the first stage, the stimulus sequence corresponding to letter presentations was organized into distinct blocks and convolved with the hemodynamic response function (HRF). A general linear model (GLM) was then applied to identify brain regions showing significantly higher or lower activity in response to the presentation of the letter “M” compared to the control condition (“W”). To minimize the effects of head motion on the results, six motion parameters derived from the preprocessing stage were included in the GLM.

In the second stage, several steps were followed. First, for each experimental condition (HC, Real, and Sham), one-sample t-tests were performed to examine the overall brain activation patterns elicited by the dual-choice oddball task. Second, a one-way ANOVA was conducted to assess significant differences in brain activation patterns across the three experimental conditions. Post-hoc tests were performed to further explore pairwise comparisons between the conditions. Lastly, brain activation values were extracted from regions showing significant differences between the three groups, and detailed charts were generated to visually present these differences.

For the behavioral results of the dual-choice oddball task, including the accuracy difference between deviant and standard stimuli (deviant - standard) and the reaction time difference between deviant and standard stimuli (deviant - standard), a one-way ANOVA was conducted to compare the three groups (HC, Real, and Sham). The significance threshold was set at p < 0.05, with FDR correction applied for multiple comparisons. Post-hoc pairwise comparisons were corrected using the Bonferroni method, with the adjusted significance threshold set at 0.05.

For the neuroimaging data of the dual-choice oddball task, multiple comparison corrections were also applied using the FDR standard. For F-tests comparing activations across the three groups, as well as post-hoc pairwise comparisons, the statistical threshold was set at pFDR < 0.05.

Throughout the intervention phase, two participants in the Real group and two in the Sham group did not complete the treatment. Additionally, issues with data quality necessitated further exclusions: excessive head motion during fMRI scanning resulted in the exclusion of one participant from the Real group, two from the Sham group, and two from the Healthy Control group. Consequently, the analysis was conducted with a final sample of 141 participants, distributed as follows: 47 in the Real group, 46 in the Sham group, and 48 in the HC group. Comparisons between the three groups (Real, Sham, and HC) revealed no significant differences in age, gender, handedness, or frame-wise displacement. However, the HC group had significantly longer years of education compared to the schizophrenia (SZ) groups (p < 0.001). Furthermore, comparisons between the two SZ groups (Real and Sham) indicated no significant differences in duration of illness or in Positive and Negative Syndrome Scale (PANSS) scores, including the Positive score, Negative score, General score, and the Total score. Detailed demographic and clinical information for each group can be found in Table 1 .

Participants’ performance was assessed using the dual-choice oddball task, measuring both reaction times (RT) and accuracy for standard and deviant stimuli. The healthy control (HC) group had shorter RTs for both standard stimuli (462.72 ± 31.81 ms) and deviant stimuli (504.82 ± 29.65 ms), with the smallest RT difference of 42.10 ± 39.41 ms. In comparison, the schizophrenia group receiving sham rTMS treatment exhibited the longest RTs for both standard stimuli (536.48 ± 32.86 ms) and deviant stimuli (643.57 ± 34.33 ms), resulting in the largest RT difference of 107.09 ± 44.73 ms. The Real rTMS group displayed intermediate RTs for standard stimuli (468.03 ± 27.68 ms) and deviant stimuli (537.62 ± 27.84 ms), with an RT difference of 69.59 ± 38.43 ms. A one-way ANOVA revealed significant variation in RT differences across the three groups (F_{(2, 138)} = 29.55, p < 0.0001). Post-hoc Tukey’s HSD tests further confirmed significant differences between all groups. The Sham group differed significantly from both the HC group (mean difference = 37.50 ms, p < 0.001) and the Real rTMS group (mean difference = -64.99 ms, p < 0.001). Additionally, the Real rTMS group showed a significant difference compared to the HC group (mean difference = -27.49 ms, p = 0.004). Results are shown in Table 1 , Figure 2 .

In terms of accuracy, the HC group performed with an average accuracy of 91.7% ± 4.6% for standard stimuli and 89.5% ± 4.4% for deviant stimuli. The Sham group displayed lower accuracy (86.0% ± 3.6% for standard and 71.4% ± 4.2% for deviant), while the Real group achieved 85.9% ± 5.2% accuracy for standard and 77.2% ± 4.6% for deviant stimuli. The accuracy difference (M - W) followed a similar pattern, with the Sham group showing the largest discrepancy (-0.146 ± 0.054), followed by the Real group (-0.087 ± 0.070) and the HC group (-0.022 ± 0.059). There was a highly significant effect of group on the accuracy difference (M - W), as indicated by ANOVA (F_{(2, 138)} = 48.96, p < 0.0001). Post-hoc Tukey’s tests revealed significant differences between all three groups. The Sham group had a significantly larger accuracy discrepancy compared to both the HC (mean difference = -0.1247, p < 0.001) and Real groups (mean difference = 0.0599, p < 0.001). The Real group also differed significantly from the HC group (mean difference = -0.0648, p < 0.001). Results are shown in Table 1 , Figure 1 .

Figure 3 presents the results of the one-sample t-test, highlighting brain activation patterns in response to deviant stimuli compared to standard stimuli. The activation patterns were largely consistent across all groups, with key regions including the bilateral middle frontal gyrus, bilateral supplementary motor area, bilateral inferior frontal gyrus, left cingulate gyrus, right insula, and bilateral inferior parietal lobule.

Figure 4 , Table 2 show the ANOVA results, revealing significant differences in brain responses among the three groups. The regions displaying these differences included the cerebellum, right insula, left thalamus, precuneus, superior temporal gyrus, and middle temporal gyrus. Post-hoc analyses, as illustrated in Figure 5 , further identified specific group differences in the cerebellum, insula, and thalamus. The Sham group showed significantly reduced activity in these regions, while the Real group demonstrated a recovery in activation levels, approaching those of the healthy control group.

A different pattern emerged in the middle temporal gyrus, precuneus, and superior temporal gyrus, as shown in Figure 6 . These areas typically exhibit negative activation during the task under normal conditions. In the Sham group, this negative activation was reduced. However, after real rTMS treatment, negative activation in these regions was restored and even surpassed that of the healthy controls, reaching its peak level.

A significant positive correlation was observed between increased activation in the right insula and improvements in accuracy in the Real-rTMS group (r = 0.45, p < 0.01), besides, a positive correlation was also found between the increased de-activation within the middle temporal gyrus and the improvements of reaction time in the Real-rTMS group (r = 0.37, p < 0.01, Figure 7 ).