Hippocampus Radiomic Biomarkers for the Diagnosis of Amnestic Mild Cognitive Impairment: A Machine Learning Method

Qi Feng, Qiaowei Song, Mei Wang, PeiPei Pang, Zhengluan Liao, Hongyang Jiang, Dinggang Shen, Zhongxiang DingFrontiers in Aging Neuroscience 2019 · DOI

Abstract

Background: Recent evidence suggests the presence of hippocampal neuroanatomical abnormalities in subjects of amnestic mild cognitive impairment (aMCI). Our study aimed to identify the radiomic biomarkers of the hippocampus for building the classification models in aMCI diagnosis. Methods: For this target, we recruited 42 subjects with aMCI and 44 normal controls (NC). The right and left hippocampi were segmented for each subject using an efficient learning-based method. Then, the radiomic analysis was applied to calculate and select the radiomic features. Finally, two logistic regression models were built based on the selected features obtained from the right and left hippocampi. Results: There were 385 features derived after calculation, and four features remained after feature selection from each group of data. The area under the receiver operating characteristic (ROC) curve, specificity, sensitivity, positive predictive value, negative predictive value, precision, recall, and F-score of the classification evaluation index of the right hippocampus logistic regression model were 0.76, 0.71, 0.69, 0.69, 0.71, 0.69, 0.69, and 0.69, and those of the left hippocampus model were 0.79, 0.71, 0.54, 0.64, 0.63, 0.64, 0.54, and 0.58, respectively. Conclusion: Results demonstrate the potential hippocampal radiomic biomarkers are valid for the aMCI diagnosis. The MRI-based radiomic analysis, with further improvement and validation, can be used to identify patients with aMCI and guide the individual treatment.

Introduction

Amnestic mild cognitive impairment (aMCI) is widely recognized as the prodromal or "pre-clinical" stage of Alzheimer’s Disease (AD). While individuals with aMCI do not yet meet the clinical criteria for dementia, they experience significant episodic memory loss that exceeds what is expected for their age. Identifying these patients early is critical, as the aMCI phase offers a vital window for interventions that may delay the progression to full-blown AD.

Traditionally, neuroimaging analysis has focused on macro-structural changes, specifically atrophy or the visible shrinking of the hippocampus. However, visual inspection by clinicians often fails to detect the subtle, high-dimensional changes in tissue texture and voxel distribution that occur before significant volume loss. This study addresses the need for objective, automated, and quantitative biomarkers by leveraging radiomics—a method that treats medical images as data to extract features invisible to the human eye.

Methods and Materials

Study Cohort

The researchers recruited 86 right-handed individuals from the Memory Clinic of Zhejiang Provincial People’s Hospital. The final cohort consisted of 42 aMCI patients (diagnosed via memory complaints and MMSE scores between 24–27) and 44 normal controls (NC) with MMSE scores ≥ 28. All participants underwent 3D T1-weighted MRI scanning using a 3.0T GE Discovery MR750 scanner.

Flowchart of subject recruitment and exclusion criteria.

Radiomics Workflow

The study employed a rigorous machine learning pipeline to identify predictive biomarkers:

  1. Segmentation: The left and right hippocampi were automatically segmented from the MRI scans using an efficient learning-based Region of Interest (ROI) method.
  2. Feature Extraction: Using the "Artificial Intelligence Kit" (A.K) software, 385 quantitative features were calculated for each subject. These included histogram features (voxel intensity distributions), form factors (3D shape and size), and texture features (complex spatial patterns like GLCM and Haralick matrices).
  3. Feature Selection: To prevent overfitting and the "curse of dimensionality," a three-step selection process was used: statistical filtering (t-tests), correlation analysis (to remove redundant features), and LASSO (Least Absolute Shrinkage and Selection Operator) regression to identify the most impactful features.
  4. Modeling: Two separate logistic regression models were developed—one for the left hippocampus and one for the right.

Results

The feature selection process successfully reduced the initial 385 features down to just four key biomarkers for each side of the hippocampus. The resulting models demonstrated moderate to strong diagnostic performance:

  • Left Hippocampus Model: Achieved an AUC of 0.79, with a specificity of 0.71 and sensitivity of 0.54.
  • Right Hippocampus Model: Achieved an AUC of 0.76, with a specificity of 0.71 and sensitivity of 0.69.

Key predictive features included "Maximum 3D Diameter" (a shape metric) and "Uniformity" (a histogram metric). Interestingly, the left hippocampus model also relied on "Inverse Difference Moment," a texture feature that describes the local homogeneity of the tissue.

Discussion

A significant finding of this study is that the left hippocampus model slightly outperformed the right. This asymmetry aligns with existing neurological research suggesting that the left hemisphere may undergo earlier or more pronounced structural changes during the early stages of cognitive decline. Furthermore, the inclusion of texture features like "Uniformity" suggests that the internal composition of the hippocampus changes in ways that are just as diagnostically relevant as its overall size.

While the results are promising, the study has limitations. The sample size of 86 is relatively small, and the sensitivity for the left hippocampus model (0.54) indicates that it missed a significant portion of aMCI cases. Future research should focus on larger, multi-center datasets and the integration of other imaging modalities, such as functional MRI (fMRI) or Diffusion Tensor Imaging (DTI), to improve sensitivity.

Conclusion

This study demonstrates that MRI-based radiomic analysis of the hippocampus can provide valid biomarkers for the diagnosis of aMCI. By moving beyond simple volume measurements to high-dimensional texture and shape analysis, machine learning models can assist clinicians in identifying patients at risk for Alzheimer's Disease earlier and more accurately, potentially guiding more effective individual treatment plans.

Summary updated

Abstract

<b>Background</b>: Recent evidence suggests the presence of hippocampal neuroanatomical abnormalities in subjects of amnestic mild cognitive impairment (aMCI). Our study aimed to identify the radiomic biomarkers of the hippocampus for building the classification models in aMCI diagnosis. <b>Methods</b>: For this target, we recruited 42 subjects with aMCI and 44 normal controls (NC). The right and left hippocampi were segmented for each subject using an efficient learning-based method. Then, the radiomic analysis was applied to calculate and select the radiomic features. Finally, two logistic regression models were built based on the selected features obtained from the right and left hippocampi. <b>Results</b>: There were 385 features derived after calculation, and four features remained after feature selection from each group of data. The area under the receiver operating characteristic (ROC) curve, specificity, sensitivity, positive predictive value, negative predictive value, precision, recall, and <i>F</i>-score of the classification evaluation index of the right hippocampus logistic regression model were 0.76, 0.71, 0.69, 0.69, 0.71, 0.69, 0.69, and 0.69, and those of the left hippocampus model were 0.79, 0.71, 0.54, 0.64, 0.63, 0.64, 0.54, and 0.58, respectively. <b>Conclusion</b>: Results demonstrate the potential hippocampal radiomic biomarkers are valid for the aMCI diagnosis. The MRI-based radiomic analysis, with further improvement and validation, can be used to identify patients with aMCI and guide the individual treatment.
BibTeX
@inproceedings{feng2019hippocampus,
  author = {Qi Feng and Qiaowei Song and Mei Wang and PeiPei Pang and Zhengluan Liao and Hongyang Jiang and Dinggang Shen and Zhongxiang Ding},
  title = {{Hippocampus Radiomic Biomarkers for the Diagnosis of Amnestic Mild Cognitive Impairment: A Machine Learning Method}},
  booktitle = {Frontiers in Aging Neuroscience},
  year = {2019},
  doi = {10.3389/fnagi.2019.00323},
}
Figure