2490

Radiomics Features of the Hippocampus Based on 3D-TIWI Improve the Diagnosis of Cerebral Small Vessel Disease with Cognitive Impairment
bingqin huang1, wei zheng1, ronghua mu1, kan deng2, jia kuang3, xiaoyan qin1, peng yang1, yuling feng1, yue xiao4, and xiqi zhu1
1Nanxishan Hospital of Guangxi Zhuang Autonomous Region, guilin, China, 2Philips Healthcare, China, guangzhou, China, 3The second Affiliated Hospital of Guilin Medical University, guilin, China, 4Guilin Medical University, guilin, China

Synopsis

Keywords: Blood Vessels, Aging, Cerebral Small Vessel Disease

Motivation: Tracking hippocampal radiomic changes over time may provide a biomarker to monitor disease progression and treatment response in cerebral small vessel disease (CSVD). Extending proven hippocampal radiomic methods from AD research to the study of CSVD represents a promising approach worthy of further investigation.

Goal(s): To develop a radiomics model based on 3D-T1WI images to improve the diagnosis of CSVD with cognitive impairment.

Approach: LASSO regression was used for feature selection and model construction

Results: The model attained an accuracy of 0.781, AUC of 0.818, sensitivity of 0.538, and specificity of 0.947 in distinguishing group 2 from NCs in the test sets.

Impact: Overall, our findings support the potential for hippocampal textural features to serve as neuroimaging biomarkers of CSVD, providing a useful tool to aid clinical decision-making in precision medicine.

Introduction

Cerebral small vessel disease accounts for approximately 25% of ischemic strokes (1) and most intracerebral hemorrhages (2). It is also the primary pathology linked to vascular cognitive impairment and dementia, resulting in substantial global disability (3). Despite its importance, there is currently no effective clinical strategy for preventing or treating CSVD (4). Therefore, the focus of current research has gradually shifted to studying methods for diagnosing CSVD.
The hippocampus has been established as a key region of interest in the diagnosis of Alzheimer's disease (AD) (5). Atrophy of the hippocampus on structural MRI is a well-validated biomarker for AD (6), with degree of hippocampal atrophy correlating with disease severity.Similar to AD, the hippocampus has also been implicated as an important region affected by cerebral small vessel disease (CSVD). Studies using MRI have found greater hippocampal atrophy in CSVD compared to healthy controls. This atrophy may be driven by ischemic lesions, microbleeds, and white matter changes that disrupt connectivity to and from the hippocampus. Hippocampal volume loss has been associated with worse cognitive performance and progression to dementia in CSVD. Given the known vulnerability of the hippocampus in both AD and CSVD, radiomic analysis of this region may offer insights for differential diagnosis. While AD is characterized by neurodegeneration, CSVD is primarily a vascular disorder. Radiomic features that quantify shape, texture, and intensity variations in the hippocampus could help distinguish between patterns of atrophy unique to AD versus CSVD(7). Furthermore, tracking hippocampal radiomic changes over time may provide a biomarker to monitor disease progression and treatment response in CSVD. Extending proven hippocampal radiomic methods from AD research to the study of CSVD represents a promising approach worthy of further investigation.

Methods

The study included 144 participants in group 1 with Montreal Cognitive Assessment (MoCA) scores between 18-26, 62 participants in group 2 with MoCA scores below 17, and 98 normal control (NC) participants with MoCA scores greater than 27 were recruited. All subjects underwent MRI including 3D T1-weighted imaging (T1WI). We proposed a semantic-aware normalization and whitening framework with a 2D UNet to automatically segment the hippocampus from the 3D T1WI scans. Radiomic features were extracted from the segmented hippocampal volumes using pyradiomics.Feature preprocessing included replacing missing values with the mean and stratified random sampling to divide subjects into training (80%) and testing (20%) sets, balanced across the three classes (normal, MoCA 18-26, MCI MoCA <17). Features were normalized to remove unit limitations. Least absolute shrinkage and selection operator (LASSO) regression to further select an optimal feature subset, using 5-fold cross-validation to tune hyperparameters and minimize error.The selected radiomic features was used to develop radiomics models. Performance was assessed in training and testing sets by receiver operating characteristic (ROC) curve analysis. Area under the ROC curve (AUC), sensitivity, and specificity were calculated to evaluate the model's ability to discriminate between the three classes. Decision curve analysis (DCA) assessed the clinical utility of the model for distinguishing MCI patients from normal controls.

Results

A total of 1197 hippocampal texture features were extracted. The model achieved an accuracy of 0.625, AUC of 0.593 with 95% CI of 0.423-0.762, sensitivity of 0.828 and specificity of 0.316 in distinguishing group 1 from NCs. The model attained accuracy of 0.683, AUC of 0.661, 95% CI of 0.489-0.833, sensitivity of 0.167, and specificity of 0.897 in distinguishing group 1 from group 2.The model attained an accuracy of 0.781, AUC of 0.818, 95% CI of 0.673-0.962, sensitivity of 0.538, and specificity of 0.947 in distinguishing group 2 from NCs.

Conclusion

In conclusion, we found that hippocampal radiomic features can distinguish patients with cognitive impairment from normal controls. Furthermore, this study demonstrates the moderately successful diagnostic classification of moderate to severe cognitive impairment versus NCs based on hippocampal radiomic features. Overall, our findings support the potential for hippocampal textural features to serve as neuroimaging biomarkers of CSVD, providing a useful tool to aid clinical decision-making in precision medicine.

Acknowledgements

No acknowledgements found.

References

  1. Feigin VL, Roth GA, Naghavi M, et al. Global burden of stroke and risk factors in 188 countries, during 1990-2013: a systematic analysis for the Global Burden of Disease Study 2013. Lancet Neurol. 2016;15(9):913-924.
  2. Pasi M, Sugita L, Xiong L, et al. Association of Cerebral Small Vessel Disease and Cognitive Decline After Intracerebral Hemorrhage. Neurology. 2021;96(2):e182-e192.
  3. Markus HS, van Der Flier WM, Smith EE, et al. Framework for Clinical Trials in Cerebral Small Vessel Disease (FINESSE): A Review. JAMA Neurol. 2022;79(11):1187-1198.
  4. Smith EE, Markus HS. New Treatment Approaches to Modify the Course of Cerebral Small Vessel Diseases. Stroke. 2020;51(1):38-46.
  5. Katabathula S, Wang Q, Xu R. Predict Alzheimer's disease using hippocampus MRI data: a lightweight 3D deep convolutional network model with visual and global shape representations. Alzheimers Res Ther. 2021;13(1):104. Published 2021 May 24.
  6. Contador J, Pérez-Millan A, Guillen N, et al. Baseline MRI atrophy predicts 2-year cognitive outcomes in early-onset Alzheimer's disease. J Neurol. 2022;269(5):2573-2583.
  7. Du Y, Zhang S, Fang Y, et al. Radiomic Features of the Hippocampus for Diagnosing Early-Onset and Late-Onset Alzheimer's Disease. Front Aging Neurosci. 2022;13:789099. Published 2022 Jan 26.

Figures

Figure 1. A–C A, group 1 vs NC. B, group 2 vs NC. C, group 1 vs group 2.NC, Montreal Cognitive Assessment score of 26 or above ;group 1，Montreal Cognitive Assessment score 18-25;group 2, Montreal Cognitive Assessment score below 17.The histogram of the Rad-signature shows the texture features selected using the least absolute shrinkage and selection operator (LASSO) method.

Figure 2.The ROC of the group 1 vs NC s in the training and test sets(A).The ROC of the group 2 vs NC s in the training and test sets(B).The ROC of the group 1 vs group 2 in the training and test sets(C).ROC, receiver operating characteristic.NC, Montreal Cognitive Assessment score of 26 or above ;group 1，Montreal Cognitive Assessment score 18-25;group 2, Montreal Cognitive Assessment score below 17.

Figure 3.The DCA of the group 1 vs NC s in the training and test sets(A).The DCA of the group 2 vs NC s in the training and test sets(B).The DCA of the group 1 vs group 2 in the training and test sets(C).DCA, decision curve analysis. The x-axis represents the threshold probability, and the y-axis represents the net benefit.

Proc. Intl. Soc. Mag. Reson. Med. 32 (2024)
2490
DOI: https://doi.org/10.58530/2024/2490