Improving the Accuracy of Brain Image Analysis in Patients with Idiopathic Normal Pressure Hydrocephalus Using Customized Tissue Probability Maps and Templates

Academic Background

Idiopathic Normal Pressure Hydrocephalus (iNPH) is a common neurological disorder characterized by gait disturbance, cognitive decline, and urinary incontinence. One of the neuroradiological features of iNPH is “Disproportionately Enlarged Subarachnoid Space Hydrocephalus” (DESH), which involves significant dilation of the ventricles and Sylvian fissures, accompanied by a narrowing of the cerebrospinal fluid (CSF) space in the superior convexity. These morphological changes pose significant challenges for the statistical analysis and segmentation of brain images in iNPH patients, especially when using traditional statistical parametric mapping software such as SPM12, which often results in brain tissue segmentation errors and inaccurate spatial normalization.

To address these issues, researchers developed the Computational Anatomy Toolbox (CAT12), which employs the Adaptive Maximum A Posteriori (AMAP) technique for brain tissue segmentation, allowing for better handling of local intensity variations. However, the default settings of CAT12 may not be suitable for all patient groups, particularly iNPH patients. Therefore, this study aimed to develop and validate customized Tissue Probability Maps (TPMs) and templates for DESH patients to improve the accuracy of brain image segmentation and spatial normalization.

Source of the Paper

This paper was authored by Shigenori Kanno and his team, with members from the Department of Behavioral Neurology and Cognitive Neuroscience at Tohoku University Graduate School of Medicine, the Department of Neurology at the University of Tokyo, the Department of Radiological Technology at South Miyagi Medical Center, and other institutions. The paper was published in 2024 in the journal Fluids and Barriers of the CNS, titled “The utility of customised tissue probability maps and templates for patients with idiopathic normal pressure hydrocephalus: a computational anatomy toolbox (CAT12) study.”

Research Process

1. Inclusion and Grouping of Study Participants

The study enrolled 298 iNPH patients and 25 healthy controls (HCs). All patients met the diagnostic criteria of the Japanese Clinical Guidelines for iNPH and underwent cranial MRI. The study was divided into two main phases: the development phase and the validation phase.

• Development Phase: Brain images from 169 out of 298 patients were selected to create customized DESH-TPMs and DESH templates. These images were processed using CAT12 for segmentation, and ultimately, images from 114 patients were used to generate the final DESH-TPM and template.
• Validation Phase: The remaining 38 patients’ images were used to validate the effectiveness of the DESH-TPM and template. Additionally, images from 25 healthy controls served as reference standards.

2. Development of Customized DESH-TPM and Template

During the development phase, researchers first segmented the brain images of 169 patients using CAT12 to generate images of gray matter, white matter, CSF, bone, and soft tissues. Subsequently, a preliminary DESH template and TPM were created using the DARTEL (Diffeomorphic Anatomical Registration Through Exponentiated Lie Algebra) algorithm. To address the morphological diversity in iNPH patients’ brains, the researchers also created a mask to expand the ventricular and Sylvian fissure regions, followed by smoothing using a Gaussian filter.

3. Validation of DESH-TPM and Template Effectiveness

In the validation phase, researchers compared the accuracy of brain segmentation and spatial normalization under three conditions: - Customized Condition: Segmentation and normalization using DESH-TPM and DESH template. - Standard Condition: Segmentation and normalization using the default settings of CAT12. - Reference Condition: Segmentation and normalization of healthy control images using the default settings of CAT12.

4. Data Analysis and Results

Researchers calculated the spatial normalization concordance rates of gray and white matter images using the 3D Structural Similarity Index (3D-SSIM) and evaluated the types of errors during segmentation. The main error types included: - Error 1: Misidentifying dura and/or extradural structures as brain tissue. - Error 2: Misidentifying white matter hypointensity (WMH) as gray matter. - Error 3: CSF image deficits.

Key Findings

1. Segmentation Error Rates

• Error 1: The error rate was 10.5% under the customized condition, 44.7% under the standard condition, and 13.6% under the reference condition. The customized condition significantly reduced the misidentification rate of dura and extradural structures.
• Error 2: The error rate was 18.4% under the customized condition, 42.1% under the standard condition, and 0% under the reference condition. The customized condition significantly improved the accuracy of WMH identification.
• Error 3: The error rate was 97.4% under the customized condition, 84.2% under the standard condition, and 28% under the reference condition. The customized condition did not significantly improve the accuracy of CSF image segmentation.

2. Spatial Normalization Accuracy

The customized condition achieved the highest concordance rates for gray and white matter image normalization, particularly in the superior convexity region. The standard condition showed significant misalignment of gray matter images in the superior convexity, while the reference condition’s normalization accuracy fell between the two.

Conclusions and Significance

This study demonstrated that using customized DESH-TPM and DESH templates significantly improved the accuracy of gray and white matter segmentation in iNPH patients, especially in the superior convexity region. However, the customized condition did not improve the accuracy of CSF image segmentation, which may be related to the inhomogeneous intensity of CSF in high-field MRI scanners. Future research needs to develop algorithms to overcome this challenge.

Research Highlights

• Innovation: This study is the first to develop customized TPMs and templates for iNPH patients, significantly improving the accuracy of brain segmentation and spatial normalization.
• Application Value: The study provides a new tool for brain image analysis in iNPH patients, aiding in more accurate diagnosis and treatment evaluation.
• Limitations: The customized condition did not improve CSF image segmentation accuracy, and further algorithm optimization is needed in the future.

Additional Valuable Information

The researchers also found that when using high-field MRI scanners (e.g., 3T Siemens Magnetom Vida), the intensity of CSF images in the superior convexity region was significantly lower, while it was higher in the infratentorial regions. This intensity inhomogeneity may be a major cause of CSF image segmentation errors. Future research will focus on developing new algorithms to address this issue.

---

Through this study, researchers have provided new solutions for brain image analysis in iNPH patients. Despite some remaining challenges, this achievement lays an important foundation for future research and clinical applications.