Hippocampus Subfields Volume Changes of Pre- and Post-treatment in Female Schizophrenia Patients
Poster No:
563
Submission Type:
Late-Breaking Abstract Submission
Authors:
Qingfeng Li1, Yingying Tang1, Chunbo Li1
Institutions:
1Shanghai Mental Health Center, Shanghai Jiao Tong University School of Medicine, Shanghai, China
First Author:
Qingfeng Li
Shanghai Mental Health Center, Shanghai Jiao Tong University School of Medicine
Shanghai, China
Shanghai Mental Health Center, Shanghai Jiao Tong University School of Medicine
Shanghai, China
Co-Author(s):
Yingying Tang
Shanghai Mental Health Center, Shanghai Jiao Tong University School of Medicine
Shanghai, China
Shanghai Mental Health Center, Shanghai Jiao Tong University School of Medicine
Shanghai, China
Chunbo Li
Shanghai Mental Health Center, Shanghai Jiao Tong University School of Medicine
Shanghai, China
Shanghai Mental Health Center, Shanghai Jiao Tong University School of Medicine
Shanghai, China
Introduction:
Finding a clinically useful neuroimaging biomarker capable of predicting treatment response in patients with schizophrenia remains a challenge. Previous studies have suggested that the hippocampus in patients with schizophrenia may be linked to treatment outcomes (Lu et al., 2024; Roeske et al., 2024). The primary aim of this study is to further investigate whether the volume of hippocampal subfields can predict the extent of symptom remission in female patients with schizophrenia following clinical treatment.
Methods:
This study included 86 female patients with schizophrenia who were hospitalized and received treatment at the Shanghai Mental Health Center (SMHC) between July 2021 and March 2023. The study was approved by the Ethics Committee of the SMHC.
All participants underwent a comprehensive assessment conducted by two experienced psychiatrists both at the time of admission and before discharge. Demographic information, including age, was recorded, along with the Positive and Negative Syndrome Scale (PANSS) scores, which assess psychotic symptoms. To measure treatment outcomes, we evaluated the patients' total PANSS scores as well as scores for each dimension from admission to discharge.
High-resolution T1-weighted structural brain images of all participants were acquired using a Siemens 3.0 Tesla Verio scanner. Automatic segmentation of hippocampal subfields and total intracranial volume (TIV) was performed using FreeSurfer v7.4.0. The final segmentation results included 19 non-overlapping regions (Iglesias et al., 2015) (Figure 1). The relative volume ratio for the anatomical regions of interest was calculated by dividing the volume in native space (ml) by TIV.
Data distribution normality was assessed with the Kolmogorov–Smirnov test. We calculated the correlation between hippocampal subfield volumes at admission and treatment outcomes (Pearson partial correlation, with age, and treatment duration as covariates, and FDR correction applied separately for the left and right brain) to assess the relationship between baseline hippocampal subfield volumes and treatment response. Using the volumes of bilateral hippocampal subfields as dependent variables, we built XGBoost regression models to predict the treatment outcome for each subject, and evaluated models using leave-one-out cross-validation (LOOCV) and grid search was used to determine the model's hyperparameters. Finally, we calculated the Pearson correlation between the predicted results of the XGBoost model and the actual treatment outcomes and analyzed the significance of the model's predictions for each treatment outcome dimension. All statistical analyses were performed with Python 3.7. p-values of ≤0.05 were considered statistically significant.
All participants underwent a comprehensive assessment conducted by two experienced psychiatrists both at the time of admission and before discharge. Demographic information, including age, was recorded, along with the Positive and Negative Syndrome Scale (PANSS) scores, which assess psychotic symptoms. To measure treatment outcomes, we evaluated the patients' total PANSS scores as well as scores for each dimension from admission to discharge.
High-resolution T1-weighted structural brain images of all participants were acquired using a Siemens 3.0 Tesla Verio scanner. Automatic segmentation of hippocampal subfields and total intracranial volume (TIV) was performed using FreeSurfer v7.4.0. The final segmentation results included 19 non-overlapping regions (Iglesias et al., 2015) (Figure 1). The relative volume ratio for the anatomical regions of interest was calculated by dividing the volume in native space (ml) by TIV.
Data distribution normality was assessed with the Kolmogorov–Smirnov test. We calculated the correlation between hippocampal subfield volumes at admission and treatment outcomes (Pearson partial correlation, with age, and treatment duration as covariates, and FDR correction applied separately for the left and right brain) to assess the relationship between baseline hippocampal subfield volumes and treatment response. Using the volumes of bilateral hippocampal subfields as dependent variables, we built XGBoost regression models to predict the treatment outcome for each subject, and evaluated models using leave-one-out cross-validation (LOOCV) and grid search was used to determine the model's hyperparameters. Finally, we calculated the Pearson correlation between the predicted results of the XGBoost model and the actual treatment outcomes and analyzed the significance of the model's predictions for each treatment outcome dimension. All statistical analyses were performed with Python 3.7. p-values of ≤0.05 were considered statistically significant.
Results:
In the left hemisphere, the hippocampal subfield that significantly correlated with the change in PANSS P4 was the HATA region (FDR-corrected p = 0.046). The region that significantly correlated with the change in the total PANSS P score was the left HATA (FDR-corrected p = 0.042).
In the right hemisphere, the hippocampal subfields that significantly correlated with the change in PANSS P4 were the fimbria (FDR-corrected p = 0.037) and the HATA (FDR-corrected p = 0.038).
The results from the XGBoost regression model indicated a significant correlation between the predicted and actual values for the changes in PANSS N4, PANSS G8, PANSS P, PANSS G14, PANSS G12, PANSS GP, PANSS G13, and PANSS T (Figures 2A, 2B). Among these, the highest correlation was observed for the change in PANSS T (r = 0.529).
In the right hemisphere, the hippocampal subfields that significantly correlated with the change in PANSS P4 were the fimbria (FDR-corrected p = 0.037) and the HATA (FDR-corrected p = 0.038).
The results from the XGBoost regression model indicated a significant correlation between the predicted and actual values for the changes in PANSS N4, PANSS G8, PANSS P, PANSS G14, PANSS G12, PANSS GP, PANSS G13, and PANSS T (Figures 2A, 2B). Among these, the highest correlation was observed for the change in PANSS T (r = 0.529).
Conclusions:
The results suggest that the volumes of the left and right HATA regions are related to the severity and improvement of symptoms in female schizophrenia patients, which may be associated with their information integration and transmission functions. Baseline bilateral hippocampal subfield volumes could serve as imaging biomarkers for predicting treatment efficacy in female patients with schizophrenia.
Disorders of the Nervous System:
Psychiatric (eg. Depression, Anxiety, Schizophrenia) 1
Modeling and Analysis Methods:
Classification and Predictive Modeling
Neuroanatomy, Physiology, Metabolism and Neurotransmission:
Subcortical Structures 2
Keywords:
Machine Learning
STRUCTURAL MRI
Sub-Cortical
Treatment
1|2Indicates the priority used for review