Machine-learning (ML) approaches have been applied for automatic diagnosis and prognosis of Alzheimer's Dementia (AD) using Magnetic Resonance Imaging (MRI). These studies use highly variable training and testing datasets and consequently confound objective comparison of classification between methodologies because of differences between datasets, feature extraction, feature selection, and validation methods. The hypothesis of this proposed project is that advanced ML co-analysis of commonly used MRI imaging and neuropsychological information may hold potential for improved hard-to-detect AD diagnosis. This study proposes a fully automated two-step AD diagnosis framework for patients with dementia through the following Specific Aims:
Aim 1 : To discover volumetric MRI imaging biomarkers for dementia by studying feature-based multiresolution-fractal texture extraction, Kullback-Leibler Divergence (KLD) multiclass feature selection, and fusion with feature-less deep ML methods. In this Aim, we will design hand-crafted MRI features and featureless deep learning methods to obtain imaging features. Multi-class KLD will be developed to select and fuse different features to generate MRI biomarkers.
Aim 2 : To combine advanced regression based feature fusion and prediction for MRI biomarkers and neuropsychological information informant history, AD risk factors, and functional status for accurate and hard-to-detect AD classification.
This Aim systematically fuses the newly discovered MRI imaging biomarkers with a battery of non-imaging features to target the atypical AD cases where volumetry and neuropsychological testing alone may not yield AD detection. This will help to reduce the number of patients that will need referral for further more expensive PET imaging for diagnosis of AD.
There is need to understand the effect of combining feature-guided deep Magnetic Resonance Imaging with other data, such as neurophysiological data, informant history, Alzheimer's dementia risk factors, and functional status, for accurate diagnosis of atypical Alzheimer's Dementia cases. Early intervention has the potential to improve the quality of life of such patients. Diagnosis at the early stage may also improve patient access to support services and enable effective planning for future treatment.
| Vidyaratne, L; Alam, M; Shboul, Z et al. (2018) Deep Learning and Texture-Based Semantic Label Fusion for Brain Tumor Segmentation. Proc SPIE Int Soc Opt Eng 2018: |
| Shboul, Zeina; Vidyaratne, Lasitha; Alam, Mahbubul et al. (2018) Glioblastoma and Survival Prediction. Brainlesion (2017) 10670:358-368 |
| Shboul, Zeina A; Reza, Sayed M S; Iftekharuddin, Khan M (2018) Quantitative MR Image Analysis for Brian Tumor. VipIMAGE 2017 (2017) 27:10-18 |
| Pei, Linmin; Reza, Syed M S; Li, Wei et al. (2017) Improved brain tumor segmentation by utilizing tumor growth model in longitudinal brain MRI. Proc SPIE Int Soc Opt Eng 10134: |
| Maier, Oskar; Menze, Bjoern H; von der Gablentz, Janina et al. (2017) ISLES 2015 - A public evaluation benchmark for ischemic stroke lesion segmentation from multispectral MRI. Med Image Anal 35:250-269 |
| Reza, Syed M S; Iftekharuddin, Khan M (2016) Glioma Grading Using Cell Nuclei Morphologic Features in Digital Pathology Images. Proc SPIE Int Soc Opt Eng 9785: |
| Bron, Esther E; Smits, Marion; van der Flier, Wiesje M et al. (2015) Standardized evaluation of algorithms for computer-aided diagnosis of dementia based on structural MRI: the CADDementia challenge. Neuroimage 111:562-79 |
| Reza, Syed M S; Mays, Randall; Iftekharuddin, Khan M (2015) Multi-fractal Detrended Texture Feature for Brain Tumor Classification. Proc SPIE Int Soc Opt Eng 9414: |
