Classification of Free-Living Body Posture with ECG Patch Accelerometers: Application to the Multicenter AIDS Cohort Study.

Purpose: As health studies increasingly monitor free-living heart performance via ECG patches with accelerometers, researchers will seek to investigate cardio-electrical responses to physical activity and sedentary behavior, increasing demand for fast, scalable methods to process accelerometer data. We extend a posture classification algorithm for accelerometers in ECG patches when researchers do not have ground-truth labels or other reference measurements (i.e., upright measurement). Methods: Men living with and without HIV in the Multicenter AIDS Cohort study wore the Zio XT® for up to two weeks (n = 1,250). Our novel extensions for posture classification include (1) estimation of an upright posture for each individual without a reference upright measurement; (2) correction of the upright estimate for device removal and re-positioning using novel spherical change-point detection; and (3) classification of upright and recumbent periods using a clustering and voting process rather than a simple inclination threshold used in other algorithms. As no posture labels exist in the free-living environment, we perform numerous sensitivity analyses and evaluate the algorithm against labelled data from the Towson Accelerometer Study, where participants wore accelerometers at the waist. Results: On average, 87.1% of participants were recumbent at 4am and 15.5% were recumbent at 1pm. Participants were recumbent 54 minutes longer on weekends compared to weekdays. Performance was good in comparison to labelled data in a separate, controlled setting (accuracy = 96.0%, sensitivity = 97.5%, specificity = 95.9%). Conclusions: Posture may be classified in the free-living environment from accelerometers in ECG patches even without measuring a standard upright position. Furthermore, algorithms that fail to account for individuals who rotate and re-attach the accelerometer may fail in the free-living environment.

Identifying Unique Subgroups of Individuals With Stroke Using Heart Rate and Steps to Characterize Physical Activity.

Background Low physical activity (PA) is associated with poor health outcomes after stroke. Step counts are a common metric of PA; however, other physiologic signals (eg, heart rate) may help to identify subgroups of individuals poststroke at varying levels of risk of poor health outcomes. Here, we aimed to identify clinically relevant subgroups of individuals poststroke based on PA profiles that leverage multiple data sources, including step count and heart rate data, from wearable devices. Methods and Results Seventy individuals poststroke participated. Participants wore a Fitbit Inspire 2 for 1 year and completed clinical assessments. We defined a group-based steps-per-minute threshold and an individual heart rate threshold to categorize each minute of PA into 1 of 4 states: high steps/high heart rate, low steps/low heart rate, high steps/low heart rate, and low steps/high heart rate. We used the proportion of time spent in each state along with steps per day, sedentary time, mean steps among minutes with high steps and high heart rate, and resting heart rate in a k-means clustering algorithm to identify subgroups and compared Activity Measure for Post-Acute Care Mobility T Score, Stroke Impact Scale, and gait speed among subgroups. We identified 3 subgroups, Active (n=8), Sedentary (n=29), and Deconditioned (n=33), which differed significantly on all clustering variables except resting heart rate. We observed significant differences in Activity Measure for Post-Acute Care Mobility T scores between subgroups, with the Deconditioned subgroup exhibiting the lowest score. Conclusions Quantifying PA with heart rate and step count using readily available wearable devices can identify clinically meaningful subgroups of individuals poststroke.

Quantifying the Association between Objectively Measured Physical Activity and Multiple Sclerosis in the UK Biobank.

INTRODUCTION: Objectively measured physical activity (PA) data were collected in the accelerometry substudy of the UK Biobank. UK Biobank also contains information about multiple sclerosis (MS) diagnosis at the time of and after PA collection. This study aimed to 1) quantify the difference in PA between prevalent MS cases and matched healthy controls, and 2) evaluate the predictive performance of objective PA measures for incident MS cases. METHODS: The first analysis compared eight accelerometer-derived PA summaries between MS patients ( N = 316) and matched controls (30 controls for each MS case). The second analysis focused on predicting time to MS diagnosis among participants who were not diagnosed with MS. A total of 19 predictors including eight measures of objective PA were compared using Cox proportional hazards models (number of events = 47; 585,900 person-years of follow-up). RESULTS: In the prevalent MS study, the difference between MS cases and matched controls was statistically significant for all PA summaries ( P < 0.001). In the incident MS study, the most predictive variable of progression to MS in univariate Cox regression models was lower age ( C = 0.604), and the most predictive PA variable was lower relative amplitude (RA, C = 0.594). A two-stage forward selection using Cox regression resulted in a model with concordance C = 0.693 and four predictors: age ( P = 0.015), stroke ( P = 0.009), Townsend deprivation index ( P = 0.874), and RA ( P = 0.004). A model including age, stroke, and RA had a concordance of C = 0.691. CONCLUSIONS: Objective PA summaries were significantly different and consistent with lower activity among study participants who had MS at the time of the accelerometry study. Among individuals who did not have MS, younger age, stroke history, and lower RA were significantly associated with a higher risk of a future MS diagnosis.

Evaluating the prediction performance of objective physical activity measures for incident Parkinson's disease in the UK Biobank.

BACKGROUND: Parkinson's disease (PD) is the fastest-growing neurological condition with over 10 million cases worldwide. While age and sex are known predictors of incident PD, there is a need to identify other predictors. This study compares the prediction performance of accelerometry-derived physical activity (PA) measures and traditional risk factors for incident PD in the UK Biobank. METHODS: The study population consisted of 92,352 UK Biobank participants without PD at baseline (43.8% male, median age 63 years with interquartile range 43-69). 245 participants were diagnosed with PD by April 1, 2021 (586,604 person-years of follow-up). The incident PD prediction performances of 10 traditional predictors and 8 objective PA measures were compared using single- and multi-variable Cox models. Prediction performance was assessed using a novel, stable statistic: the repeated cross-validated concordance (rcvC). Sensitivity analyses were conducted where PD cases diagnosed within the first six months, one year, and two years were deleted. RESULTS: Single-predictor Cox regression models indicated that all PA measures were statistically significant (p-values < 0.0001). The highest-performing individual predictors were total acceleration (TA) (rcvC = 0.813) among PA measures, and age (rcvC = 0.757) among traditional predictors. The two-step forward-selection process produced a model containing age, sex, and TA (rcvC = 0.851). Adding TA to the model increased the rcvC by 9.8% (p-value < 0.0001). Results were largely unchanged in sensitivity analyses. CONCLUSIONS: Objective PA summaries have better single-predictor model performance than known risk factors and increase the prediction performance substantially when added to models with age and sex.

MISPEL: A supervised deep learning harmonization method for multi-scanner neuroimaging data.

Large-scale data obtained from aggregation of already collected multi-site neuroimaging datasets has brought benefits such as higher statistical power, reliability, and robustness to the studies. Despite these promises from growth in sample size, substantial technical variability stemming from differences in scanner specifications exists in the aggregated data and could inadvertently bias any downstream analyses on it. Such a challenge calls for data normalization and/or harmonization frameworks, in addition to comprehensive criteria to estimate the scanner-related variability and evaluate the harmonization frameworks. In this study, we propose MISPEL (Multi-scanner Image harmonization via Structure Preserving Embedding Learning), a supervised multi-scanner harmonization method that is naturally extendable to more than two scanners. We also designed a set of criteria to investigate the scanner-related technical variability and evaluate the harmonization techniques. As an essential requirement of our criteria, we introduced a multi-scanner matched dataset of 3T T1 images across four scanners, which, to the best of our knowledge is one of the few datasets of this kind. We also investigated our evaluations using two popular segmentation frameworks: FSL and segmentation in statistical parametric mapping (SPM). Lastly, we compared MISPEL to popular methods of normalization and harmonization, namely White Stripe, RAVEL, and CALAMITI. MISPEL outperformed these methods and is promising for many other neuroimaging modalities.

Fast Multilevel Functional Principal Component Analysis.

We introduce fast multilevel functional principal component analysis (fast MFPCA), which scales up to high dimensional functional data measured at multiple visits. The new approach is orders of magnitude faster than and achieves comparable estimation accuracy with the original MFPCA (Di et al., 2009). Methods are motivated by the National Health and Nutritional Examination Survey (NHANES), which contains minute-level physical activity information of more than 10000 participants over multiple days and 1440 observations per day. While MFPCA takes more than five days to analyze these data, fast MFPCA takes less than five minutes. A theoretical study of the proposed method is also provided. The associated function mfpca.face() is available in the R package refund.

Cardiovascular mortality risk prediction using objectively measured physical activity phenotypes in NHANES 2003-2006.

Increased physical activity (PA) has been associated with a decreased risk of cardiovascular disease (CVD) and mortality. However, most previous studies use self-reported PA instead of objectively measured PA assessed by wearable accelerometers. To the best of our knowledge, there have not been studies that quantified the univariate and multivariate ability of objectively measured PA summaries to predict the risk of CVD mortality. We investigate the ability of objectively measured PA summary variables to predict CVD mortality: as individual predictors, as part of the best multivariate model incorporating traditional predictors, and as additions to the best multivariate model using only traditional CVD predictors. Data were collected in the National Health and Nutrition Examination Survey 2003-2006 waves for US participants aged 50-85. The predictive ability was measured using Concordance, sometimes referred to as the C-statistic. Specifically, we calculated 10-fold cross-validated concordance (CVC) in survey-weighted Cox proportional hazard models. The best univariate predictor of CVD mortality was total activity count (outperformed age). In multivariate models, two of the eight predictors identified using the improvement in CVC threshold of 0.001 were PA measures (CVC = 0.844). The best model without physical activity (7 predictors) had CVC of 0.830. The addition of PA measures to the best traditional model was significantly better at predicting CVD mortality (P < 0.001). Accelerometer-derived PA measures have excellent cardiovascular mortality prediction performance. Wearable accelerometers have a potential for assessment of individuals' CVD mortality risks.

Patterns of objectively measured physical activity differ between men living with and without HIV.

OBJECTIVE: To use accelerometers to quantify differences in physical activity (PA) by HIV serostatus and HIV viral load (VL) in the Multicenter AIDS Cohort Study (MACS). METHODS: MACS participants living with (PLWH, n = 631) and without (PWOH, n = 578) HIV wore an ambulatory electrocardiogram monitor containing an accelerometer for 1-14 days. PA was summarized as cumulative mean absolute deviation (MAD) during the 10 most active consecutive hours (M10), cumulative MAD during the six least active consecutive hours (L6), and daily time recumbent (DTR). PA summaries were compared by HIV serostatus and by detectability of VL (>20 vs. ≤20 copies/ml) using linear mixed models adjusted for sociodemographics, weight, height, substance use, physical function, and clinical factors. RESULTS: In sociodemographic-adjusted models, PLWH with a detectable VL had higher L6 (ß = 0.58 mg, P = 0.027) and spent more time recumbent (ß = 53 min/day, P = 0.003) than PWOH. PLWH had lower M10 than PWOH (undetectable VL ß = -1.62 mg, P = 0.027; detectable VL ß = -1.93 mg, P = 0.12). A joint test indicated differences in average PA measurements by HIV serostatus and VL (P = 0.001). However, differences by HIV serostatus in M10 and DTR were attenuated and no longer significant after adjustment for renal function, serum lipids, and depressive symptoms. CONCLUSIONS: Physical activity measures differed significantly by HIV serostatus and VL. Higher L6 among PLWH with detectable VL may indicate reduced amount or quality of sleep compared to PLWH without detectable VL and PWOH. Lower M10 among PLWH indicates lower amounts of physical activity compared to PWOH.

Comparison of Accelerometry-Based Measures of Physical Activity: Retrospective Observational Data Analysis Study.

BACKGROUND: Given the evolution of processing and analysis methods for accelerometry data over the past decade, it is important to understand how newer summary measures of physical activity compare with established measures. OBJECTIVE: We aimed to compare objective measures of physical activity to increase the generalizability and translation of findings of studies that use accelerometry-based data. METHODS: High-resolution accelerometry data from the Baltimore Longitudinal Study on Aging were retrospectively analyzed. Data from 655 participants who used a wrist-worn ActiGraph GT9X device continuously for a week were summarized at the minute level as ActiGraph activity count, monitor-independent movement summary, Euclidean norm minus one, mean amplitude deviation, and activity intensity. We calculated these measures using open-source packages in R. Pearson correlations between activity count and each measure were quantified both marginally and conditionally on age, sex, and BMI. Each measures pair was harmonized using nonparametric regression of minute-level data. RESULTS: Data were from a sample (N=655; male: n=298, 45.5%; female: n=357, 54.5%) with a mean age of 69.8 years (SD 14.2) and mean BMI of 27.3 kg/m2 (SD 5.0). The mean marginal participant-specific correlations between activity count and monitor-independent movement summary, Euclidean norm minus one, mean amplitude deviation, and activity were r=0.988 (SE 0.0002324), r=0.867 (SE 0.001841), r=0.913 (SE 0.00132), and r=0.970 (SE 0.0006868), respectively. After harmonization, mean absolute percentage errors of predicting total activity count from monitor-independent movement summary, Euclidean norm minus one, mean amplitude deviation, and activity intensity were 2.5, 14.3, 11.3, and 6.3, respectively. The accuracies for predicting sedentary minutes for an activity count cut-off of 1853 using monitor-independent movement summary, Euclidean norm minus one, mean amplitude deviation, and activity intensity were 0.981, 0.928, 0.904, and 0.960, respectively. An R software package called SummarizedActigraphy, with a unified interface for computation of the measures from raw accelerometry data, was developed and published. CONCLUSIONS: The findings from this comparison of accelerometry-based measures of physical activity can be used by researchers and facilitate the extension of knowledge from existing literature by demonstrating the high correlation between activity count and monitor-independent movement summary (and other measures) and by providing harmonization mapping.

Fast Univariate Inference for Longitudinal Functional Models.

We propose fast univariate inferential approaches for longitudinal Gaussian and non-Gaussian functional data. The approach consists of three steps: (1) fit massively univariate pointwise mixed effects models; (2) apply any smoother along the functional domain; and (3) obtain joint confidence bands using analytic approaches for Gaussian data or a bootstrap of study participants for non-Gaussian data. Methods are motivated by two applications: (1) Diffusion Tensor Imaging (DTI) measured at multiple visits along the corpus callosum of multiple sclerosis (MS) patients; and (2) physical activity data measured by body-worn accelerometers for multiple days. An extensive simulation study indicates that model fitting and inference are accurate and much faster than existing approaches. Moreover, the proposed approach was the only one that was computationally feasible for the physical activity data application. Methods are accompanied by R software, though the method is "read-and-use", as it can be implemented by any analyst who is familiar with mixed effects model software.

Fixed-effects inference and tests of correlation for longitudinal functional data.

We propose an inferential framework for fixed effects in longitudinal functional models and introduce tests for the correlation structures induced by the longitudinal sampling procedure. The framework provides a natural extension of standard longitudinal correlation models for scalar observations to functional observations. Using simulation studies, we compare fixed effects estimation under correctly and incorrectly specified correlation structures and also test the longitudinal correlation structure. Finally, we apply the proposed methods to a longitudinal functional dataset on physical activity. The computer code for the proposed method is available at https://github.com/rli20ST758/FILF.

Defining Incidence of Acute Compartment Syndrome in the Research Setting: A Proposed Method From the PACS Study.

OBJECTIVE: To compare the retrospective decision of an expert panel who assessed likelihood of acute compartment syndrome (ACS) in a patient with a high-risk tibia fracture with decision to perform fasciotomy. DESIGN: Prospective observational study. SETTING: Seven Level 1 trauma centers. PATIENTS/PARTICIPANTS: One hundred eighty-two adults with severe tibia fractures. MAIN OUTCOME MEASUREMENTS: Diagnostic performance (sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and receiver-operator curve) of an expert panel's assessment of likelihood ACS compared with fasciotomy as the reference diagnostic standard. SECONDARY OUTCOMES: The interrater reliability of the expert panel as measured by the Krippendorff alpha. Expert panel consensus was determined using the percent of panelists in the majority group of low (expert panel likelihood of ≤0.3), uncertain (0.3-0.7), or high (>0.7) likelihood of ACS. RESULTS: Comparing fasciotomy (the diagnostic standard) and the expert panel's assessment as the diagnostic classification (test), the expert panel's determination of uncertain or high likelihood of ACS (threshold >0.3) had a sensitivity of 0.90 (0.70, 0.99), specificity of 0.95 (0.90, 0.98), PPV of 0.70 (0.50, 0.86), and NPV of 0.99 (0.95, 1.00). When a threshold of >0.7 was set as a positive diagnosis, the expert panel assessment had a sensitivity of 0.67 (0.43, 0.85), specificity of 0.98 (0.95, 1.00), PPV of 0.82 (0.57, 0.96), and NPV of 0.96 (0.91, 0.98). CONCLUSION: In our study, the retrospective assessment of an expert panel of the likelihood of ACS has good specificity and excellent NPV for fasciotomy, but only low-to-moderate sensitivity and PPV. The discordance between the expert panel-assessed likelihood of ACS and the decision to perform fasciotomy suggests that concern regarding potential diagnostic bias in studies of ACS is warranted. LEVEL OF EVIDENCE: Diagnostic Level I. See Instructions for Authors for a complete description of levels of evidence.

Additive Functional Cox Model.

We propose the Additive Functional Cox Model to flexibly quantify the association between functional covariates and time to event data. The model extends the linear functional proportional hazards model by allowing the association between the functional covariate and log hazard to vary non-linearly in both the functional domain and the value of the functional covariate. Additionally, we introduce critical transformations of the functional covariate which address the weak model identifiability in areas of information sparsity and discuss their impact on interpretation and inference. We also introduce a novel estimation procedure that accounts for identifiability constraints directly during model fitting. Methods are applied to the National Health and Nutrition Examination Survey (NHANES) 2003-2006 accelerometry data and quantify new and interpretable circadian patterns of physical activity that are associated with all-cause mortality. We also introduce a simple and novel simulation framework for generating survival data with functional predictors which resemble the observed data. The accompanying inferential R software is fast, open source and publicly available. Our data application and simulations are fully reproducible through the accompanying vignette.

A multi-scanner neuroimaging data harmonization using RAVEL and ComBat.

Modern neuroimaging studies frequently combine data collected from multiple scanners and experimental conditions. Such data often contain substantial technical variability associated with image intensity scale (image intensity scales are not the same in different images) and scanner effects (images obtained from different scanners contain substantial technical biases). Here we evaluate and compare results of data analysis methods without any data transformation (RAW), with intensity normalization using RAVEL, with regional harmonization methods using ComBat, and a combination of RAVEL and ComBat. Methods are evaluated on a unique sample of 16 study participants who were scanned on both 1.5T and 3T scanners a few months apart. Neuroradiological evaluation was conducted for 7 different regions of interest (ROI's) pertinent to Alzheimer's disease (AD). Cortical measures and results indicate that: (1) RAVEL substantially improved the reproducibility of image intensities; (2) ComBat is preferred over RAVEL and the RAVEL-ComBat combination in terms of regional level harmonization due to more consistent harmonization across subjects and image-derived measures; (3) RAVEL and ComBat substantially reduced bias compared to analysis of RAW images, but RAVEL also resulted in larger variance; and (4) the larger root mean square deviation (RMSD) of RAVEL compared to ComBat is due mainly to its larger variance.

Estimation of free-living walking cadence from wrist-worn sensor accelerometry data and its association with SF-36 quality of life scores.

Objective. We evaluate the stride segmentation performance of the Adaptive Empirical Pattern Transformation (ADEPT) for subsecond-level accelerometry data collected in the free-living environment using a wrist-worn sensor.Approach. We substantially expand the scope of the existing ADEPT pattern-matching algorithm. Methods are applied to subsecond-level accelerometry data collected continuously for 4 weeks in 45 participants, including 30 arthritis and 15 control patients. We estimate the daily walking cadence for each participant and quantify its association with SF-36 quality of life measures.Main results. We provide free, open-source software to segment individual walking strides in subsecond-level accelerometry data. Walking cadence is significantly associated with the role physical score reported via SF-36 after adjusting for age, gender, weight and height.Significance. Methods provide automatic, precise walking stride segmentation, which allows estimation of walking cadence from free-living wrist-worn accelerometry data. Results provide new evidence of associations between free-living walking parameters and health outcomes.

Association of Spectral-Domain OCT With Long-term Disability Worsening in Multiple Sclerosis.

OBJECTIVE: To evaluate whether a retinal spectral-domain optical coherence tomography (SD-OCT) assessment at baseline is associated with long-term disability worsening in people with multiple sclerosis (PwMS), we performed SD-OCT and Expanded Disability Status Scale (EDSS) assessments among 132 PwMS at baseline and at a median of 10 years later. METHODS: In this prospective, longitudinal study, participants underwent SD-OCT, EDSS, and visual acuity (VA) assessments at baseline and at follow-up. Statistical analyses were performed using generalized linear regression models, adjusted for age, sex, race, multiple sclerosis (MS) subtype, and baseline disability. We defined clinically meaningful EDSS worsening as an increase of ≥2.0 if baseline EDSS score was <6.0 or an increase of ≥1.0 if baseline EDSS score was ≥6.0. RESULTS: A total of 132 PwMS (mean age 43 years; 106 patients with relapsing-remitting MS) were included in analyses. Median duration of follow-up was 10.4 years. In multivariable models excluding eyes with prior optic neuritis, relative to patients with an average baseline ganglion cell + inner plexiform layer (GCIPL) thickness ≥70 µm (the mean GCIPL thickness of all eyes at baseline), an average baseline GCIPL thickness <70 µm was associated with a 4-fold increased odds of meaningful EDSS worsening (adjusted odds ratio [OR] 3.97, 95% confidence interval [CI] 1.24-12.70; p = 0.02) and an almost 3-fold increased odds of low-contrast VA worsening (adjusted OR 2.93, 95% CI 1.40-6.13; p = 0.04). CONCLUSIONS: Lower baseline GCIPL thickness on SD-OCT is independently associated with long-term disability worsening in MS. Accordingly, SD-OCT at a single time point may help guide therapeutic decision-making among individual PwMS. CLASSIFICATION OF EVIDENCE: This study provides Class I evidence that lower baseline GCIPL thickness on SD-OCT is independently associated with long-term disability worsening in MS.

Adaptive empirical pattern transformation (ADEPT) with application to walking stride segmentation.

Quantifying gait parameters and ambulatory monitoring of changes in these parameters have become increasingly important in epidemiological and clinical studies. Using high-density accelerometry measurements, we propose adaptive empirical pattern transformation (ADEPT), a fast, scalable, and accurate method for segmentation of individual walking strides. ADEPT computes the covariance between a scaled and translated pattern function and the data, an idea similar to the continuous wavelet transform. The difference is that ADEPT uses a data-based pattern function, allows multiple pattern functions, can use other distances instead of the covariance, and the pattern function is not required to satisfy the wavelet admissibility condition. Compared to many existing approaches, ADEPT is designed to work with data collected at various body locations and is invariant to the direction of accelerometer axes relative to body orientation. The method is applied to and validated on accelerometry data collected during a $450$-m outdoor walk of $32$ study participants wearing accelerometers on the wrist, hip, and both ankles. Additionally, all scripts and data needed to reproduce presented results are included in supplementary material available at Biostatistics online.

Postcardiac arrest neurological prognostication with quantitative regional cerebral densitometry.

PURPOSE: To quantitatively assess the severity of anoxic-ischemic brain injury early after cardiac arrest (CA) using a novel automated method applied to head computed tomography (HCT). METHODS: Adult patients who were comatose and underwent HCT < 24 h after arrest were included in a retrospective analysis. Principal endpoint was unfavorable outcome (UO) defined as Cerebral Performance Category (CPC) of 3-5 at hospital discharge. We developed an automated processing algorithm for HCT images to be registered, atlas-segmented in 181 regions, and region-specific radiologic densities determined in Hounsfield Units. This approach was compared with an established manual method evaluating grey-white matter ratios (GWR). We tested univariable and multivariable prognostic models which integrated clinical and HCT features including densities in lobes and in nodes of cerebral networks linked to CA recovery. RESULTS: Ninety-one patients were enrolled among whom 66 (73%) had an UO. HCTs were interpreted as normal or without acute abnormality by a neuroradiologist in 77 cases (85%). Compared to the favorable outcome group, UO patients had significantly lower densities in all lobes and in nodes of cerebral networks. A model combining clinical variables with the automated method applied to cerebral network nodes had the highest prognostic performance although not significantly different than the combined clinical-GWR method (AUC [95% CI] 0.94 [0.86-1.00] and 0.92 [0.83-1.00] respectively). CONCLUSION: In comatose survivors of CA, automated quantitative analysis of HCT revealed very early multifocal changes in brain tissue density which are mostly overlooked on conventional neuroradiologic interpretation and are associated with neurological outcome.

Evaluating White Matter Lesion Segmentations with Refined Sørensen-Dice Analysis.

The Sørensen-Dice index (SDI) is a widely used measure for evaluating medical image segmentation algorithms. It offers a standardized measure of segmentation accuracy which has proven useful. However, it offers diminishing insight when the number of objects is unknown, such as in white matter lesion segmentation of multiple sclerosis (MS) patients. We present a refinement for finer grained parsing of SDI results in situations where the number of objects is unknown. We explore these ideas with two case studies showing what can be learned from our two presented studies. Our first study explores an inter-rater comparison, showing that smaller lesions cannot be reliably identified. In our second case study, we demonstrate fusing multiple MS lesion segmentation algorithms based on the insights into the algorithms provided by our analysis to generate a segmentation that exhibits improved performance. This work demonstrates the wealth of information that can be learned from refined analysis of medical image segmentations.