Detection of subtle white matter lesions in MRI through texture feature extraction and boundary delineation using an embedded clustering strategy

Abstract

White matter lesions (WML) underlie multiple brain disorders, and automatic WML segmentation is crucial to evaluate the natural disease course and effectiveness of clinical interventions, including drug discovery. Although recent research has achieved tremendous progress in WML segmentation, accurate detection of subtle WML present early in the disease course remains particularly challenging. Here we propose an approach to automatic WML segmentation of mild WML loads using an intensity standardisation technique, gray level co-occurrence matrix (GLCM) embedded clustering technique, and random forest (RF) classifier to extract texture features and identify morphology specific to true WML. We precisely define their boundaries through a local outlier factor (LOF) algorithm that identifies edge pixels by local density deviation relative to its neighbors. The automated approach was validated on 32 human subjects, demonstrating strong agreement and correlation (excluding one outlier) with manual delineation by a neuroradiologist through Intra-Class Correlation (ICC = 0.881, 95% CI 0.769, 0.941) and Pearson correlation (r = 0.895, p-value < 0.001), respectively, and outperforming three leading algorithms (Trimmed Mean Outlier Detection, Lesion Prediction Algorithm, and SALEM-LS) in five of the six established key metrics defined in the MICCAI Grand Challenge. By facilitating more accurate segmentation of subtle WML, this approach may enable earlier diagnosis and intervention.

Early detection of white matter hyperintensities using SHIVA-WMH detector

“…The LST-LPA is based on the logistic regression model that uses intensity and location information of FLAIR images to classify WMH. It was originally developed to segment WMH in MS patients (Schmidt, 2017a), but has been applied widely to quantify age-related WMH as well (Ribaldi et al, 2021) and has also been demonstrated to show robust performance across diverse datasets with age-related WMH (Heinen et al, 2019;Vanderbecq et al, 2020) These scores are the highest among several recent works that explicitly evaluated their methods in participants with mild WMH burden of less than 5 mL (Khademi et al, 2021;Mojiri Forooshani et al, 2022;Ong et al, 2022;Rachmadi et al, 2018). For the purpose of comparison with our tool, it also had the advantage of being trained with multisite imaging datasets that did not include the MWC dataset, allowing for a fair comparison of performance with our tool in this cohort, unlike the PGS, whose training data included MWC testing data set aside in the present study.…”

“…UKB; Alfaro-Almagro et al, 2018; ADNI-3; Gunter et al, 2017; Rhineland Study; Lohner et al, 2022). Yet, several recent works on age-related WMH detection methods that explicitly evaluated their methods in participants with mild WMH burden (< 5 mL) had not included 3D FLAIR datasets for training or evaluation (Khademi et al, 2021; Mojiri Forooshani et al, 2022; Ong et al, 2022; Rachmadi et al, 2018). Our work is unique in taking advantage of the 1mm isotropic 3D FLAIR scans from young adults and the detailed delineation of every visible WMH in every slice to train the 3D Unet-based model, with the objective to apply our tool in other cohorts with similar acquisitions to accurately describe WMH burden across the adult lifespan.…”

“…The true advantage of these advanced methods may be more evident for the segmentation of WMH in subjects with relatively mild lesion load (< 5mL), as recently suggested (Khademi et al, 2021; Li et al, 2022; Rachmadi et al, 2018). However, even those studies evaluating their method in subjects with mild WMH burden with DL-based (Khademi et al, 2021; Rachmadi et al, 2018) or other approaches (Ong et al, 2022; Rachmadi et al, 2020) primarily use databases with 2D FLAIR acquisition with the slice thickness ranging from 3 to 5 mm, which precludes detection of small WMH not on the plane of acquisition. To our knowledge, no study has explicitly optimized the segmentation performance in high-quality 3D FLAIR scans from healthy young-to middle-aged adults with very mild WMH burden.…”

Early detection of white matter hyperintensities using SHIVA‐WMH detector

“…It was chosen based on the high reported performance in cases of very mild WMH burden in a multi-cohort dataset of 50 subjects with mean WMH volume of $2 ml, with VL-and CL-Dice scores of 0.84 and 0.72, respectively (Mojiri Forooshani et al, 2022). These scores are the highest among several recent works that explicitly evaluated their methods in participants with mild WMH burden of less than 5 ml (Khademi et al, 2021;Mojiri Forooshani et al, 2022;Ong et al, 2022;Rachmadi et al, 2018). For the purpose of comparison with our tool, it also had the advantage of being trained with multisite imaging datasets that did not include the MWC dataset, allowing for a fair comparison of performance with our tool in this cohort, unlike the PGS, whose training data included MWC testing data set aside in the present study.…”

Current and Future Developments in Imaging and Treatment of White Matter Disease: A Systematic Review

“…A novel method that automates mild WML segmentation with intensity standardization is based on the similarity between WMLs and ischemia on FLAIR MRI, which is seen in MS. The use of RF classifiers with grey-level co-occurrence matrix-embedded clustering for texture and morphology recognition was the subject of one study [ 6 ]. An additional study uses provincial MRI to try to clarify the hazy connection between multiple sclerosis (MS) white matter lesions (WMLs) and cortical lesions (CLs).…”