Estimating the volume of penumbra in rodents using DTI and stack-based ensemble machine learning framework.

BACKGROUND: This study investigates the potential of diffusion tensor imaging (DTI) in identifying penumbral volume (PV) compared to the standard gadolinium-required perfusion-diffusion mismatch (PDM), utilizing a stack-based ensemble machine learning (ML) approach with enhanced explainability. METHODS: Sixteen male rats were subjected to middle cerebral artery occlusion. The penumbra was identified using PDM at 30 and 90 min after occlusion. We used 11 DTI-derived metrics and 14 distance-based features to train five voxel-wise ML models. The model predictions were integrated using stack-based ensemble techniques. ML-estimated and PDM-defined PVs were compared to evaluate model performance through volume similarity assessment, the Pearson correlation analysis, and Bland-Altman analysis. Feature importance was determined for explainability. RESULTS: In the test rats, the ML-estimated median PV was 106.4 mL (interquartile range 44.6-157.3 mL), whereas the PDM-defined median PV was 102.0 mL (52.1-144.9 mL). These PVs had a volume similarity of 0.88 (0.79-0.96), a Pearson correlation coefficient of 0.93 (p < 0.001), and a Bland-Altman bias of 2.5 mL (2.4% of the mean PDM-defined PV), with 95% limits of agreement ranging from -44.9 to 49.9 mL. Among the features used for PV prediction, the mean diffusivity was the most important feature. CONCLUSIONS: Our study confirmed that PV can be estimated using DTI metrics with a stack-based ensemble ML approach, yielding results comparable to the volume defined by the standard PDM. The model explainability enhanced its clinical relevance. Human studies are warranted to validate our findings. RELEVANCE STATEMENT: The proposed DTI-based ML model can estimate PV without the need for contrast agent administration, offering a valuable option for patients with kidney dysfunction. It also can serve as an alternative if perfusion map interpretation fails in the clinical setting. KEY POINTS: • Penumbral volume can be estimated by DTI combined with stack-based ensemble ML. • Mean diffusivity was the most important feature used for predicting penumbral volume. • The proposed approach can be beneficial for patients with kidney dysfunction.

Thalamocortical coherence predicts persistent postconcussive symptoms.

The pathogenetic mechanism of persistent post-concussive symptoms (PCS) following concussion remains unclear. Thalamic damage is known to play a role in PCS prolongation while the evidence and biomarkers that trigger persistent PCS have never been elucidated. We collected longitudinal neuroimaging and behavior data from patients and rodents after concussion, complemented with rodents' histological staining data, to unravel the early biomarkers of persistent PCS. Diffusion tensor imaging (DTI) were acquired to investigated the thalamic damage, while quantitative thalamocortical coherence was derived through resting-state functional MRI for evaluating thalamocortical functioning and predicting long-term behavioral outcome. Patients with prolonged symptoms showed abnormal DTI-derived indices at the boundaries of bilateral thalami (peri-thalamic regions). Both patients and rats with persistent symptoms demonstrated enhanced thalamocortical coherence between different thalamocortical circuits, which disrupted thalamocortical multifunctionality. In rodents, the persistent DTI abnormalities were validated in thalamic reticular nucleus (TRN) through immunohistochemistry, and correlated with enhanced thalamocortical coherence. Strong predictive power of these coherence biomarkers for long-term PCS was also validated using another patient cohort. Postconcussive events may begin with persistent TRN injury, followed by disrupted thalamocortical coherence and prolonged PCS. Functional MRI-based coherence measures can be surrogate biomarkers for early prediction of long-term PCS.

Automatic segmentation and radiomic texture analysis for osteoporosis screening using chest low-dose computed tomography.

OBJECTIVE: This study developed a diagnostic tool combining machine learning (ML) segmentation and radiomic texture analysis (RTA) for bone density screening using chest low-dose computed tomography (LDCT). METHODS: A total of 197 patients who underwent LDCT followed by dual-energy X-ray absorptiometry were analyzed. First, an autosegmentation model was trained using LDCT to delineate the thoracic vertebral body (VB). Second, a two-level classifier was developed using radiomic features extracted from VBs for the hierarchical pairwise classification of each patient's bone status. All the patients were initially classified as either normal or abnormal, and all patients with abnormal bone density were then subdivided into an osteopenia group and an osteoporosis group. The performance of the classifier was evaluated through fivefold cross-validation. RESULTS: The model for automated VB segmentation achieved a Sorenson-Dice coefficient of 0.87 ± 0.01. Furthermore, the area under the receiver operating characteristic curve scores for the two-level classifier were 0.96 ± 0.01 for detecting abnormal bone density (accuracy = 0.91 ± 0.02; sensitivity = 0.93 ± 0.03; specificity = 0.89 ± 0.03) and 0.98 ± 0.01 for distinguishing osteoporosis (accuracy = 0.94 ± 0.02; sensitivity = 0.95 ± 0.03; specificity = 0.93 ± 0.03). The testing prediction accuracy levels for the first- and second-level classifiers were 0.92 ± 0.04 and 0.94 ± 0.05, respectively. The overall testing prediction accuracy of our method was 0.90 ± 0.05. CONCLUSION: The combination of ML segmentation and RTA for automated bone density prediction based on LDCT scans is a feasible approach that could be valuable for osteoporosis screening during lung cancer screening. KEY POINTS: • This study developed an automatic diagnostic tool combining machine learning-based segmentation and radiomic texture analysis for bone density screening using chest low-dose computed tomography. • The developed method enables opportunistic screening without quantitative computed tomography or a dedicated phantom. • The developed method could be integrated into the current clinical workflow and used as an adjunct for opportunistic screening or for patients who are ineligible for screening with dual-energy X-ray absorptiometry.

Personalized Prediction of Postconcussive Working Memory Decline: A Feasibility Study.

Concussion, also known as mild traumatic brain injury (mTBI), commonly causes transient neurocognitive symptoms, but in some cases, it causes cognitive impairment, including working memory (WM) deficit, which can be long-lasting and impede a patient's return to work. The predictors of long-term cognitive outcomes following mTBI remain unclear, because abnormality is often absent in structural imaging findings. Previous studies have demonstrated that WM functional activity estimated from functional magnetic resonance imaging (fMRI) has a high sensitivity to postconcussion WM deficits and may be used to not only evaluate but guide treatment strategies, especially targeting brain areas involved in postconcussion cognitive decline. The purpose of the study was to determine whether machine learning-based models using fMRI biomarkers and demographic or neuropsychological measures at the baseline could effectively predict the 1-year cognitive outcomes of concussion. We conducted a prospective, observational study of patients with mTBI who were compared with demographically matched healthy controls enrolled between September 2015 and August 2020. Baseline assessments were collected within the first week of injury, and follow-ups were conducted at 6 weeks, 3 months, 6 months, and 1 year. Potential demographic, neuropsychological, and fMRI features were selected according to their significance of correlation with the estimated changes in WM ability. The support vector machine classifier was trained using these potential features and estimated changes in WM between the predefined time periods. Patients demonstrated significant cognitive recovery at the third month, followed by worsened performance after 6 months, which persisted until 1 year after a concussion. Approximately half of the patients experienced prolonged cognitive impairment at the 1-year follow up. Satisfactory predictions were achieved for patients whose WM function did not recover at 3 months (accuracy = 87.5%), 6 months (accuracy = 83.3%), and 1 year (accuracy = 83.3%) and performed worse at the 1-year follow-up compared to the baseline assessment (accuracy = 83.3%). This study demonstrated the feasibility of personalized prediction for long-term postconcussive WM outcomes based on baseline fMRI and demographic features, opening a new avenue for early rehabilitation intervention in selected individuals with possible poor long-term cognitive outcomes.

Identification of Differentially Expressed Genes in Different Glioblastoma Regions and Their Association with Cancer Stem Cell Development and Temozolomide Response.

The molecular heterogeneity of gene expression profiles of glioblastoma multiforme (GBM) are the most important prognostic factors for tumor recurrence and drug resistance. Thus, the aim of this study was to identify potential target genes related to temozolomide (TMZ) resistance and GBM recurrence. The genomic data of patients with GBM from The Cancer Genome Atlas (TCGA; 154 primary and 13 recurrent tumors) and a local cohort (29 primary and 4 recurrent tumors), samples from different tumor regions from a local cohort (29 tumor and 25 peritumoral regions), and Gene Expression Omnibus data (GSE84465, single-cell RNA sequencing; 3589 cells) were included in this study. Critical gene signatures were identified based an analysis of differentially expressed genes (DEGs). DEGs were further used to evaluate gene enrichment levels among primary and recurrent GBMs and different tumor regions through gene set enrichment analysis. Protein-protein interactions (PPIs) were incorporated into gene regulatory networks to identify the affected metabolic pathways. The enrichment levels of 135 genes were identified in the peritumoral regions as being risk signatures for tumor recurrence. Fourteen genes (DVL1, PRKACB, ARRB1, APC, MAPK9, CAMK2A, PRKCB, CACNA1A, ERBB4, RASGRF1, NF1, RPS6KA2, MAPK8IP2, and PPM1A) derived from the PPI network of 135 genes were upregulated and involved in the regulation of cancer stem cell (CSC) development and relevant signaling pathways (Notch, Hedgehog, Wnt, and MAPK). The single-cell data analysis results indicated that 14 key genes were mainly expressed in oligodendrocyte progenitor cells, which could produce a CSC niche in the peritumoral region. The enrichment levels of 336 genes were identified as biomarkers for evaluating TMZ resistance in the solid tumor region. Eleven genes (ARID5A, CDC42EP3, CDKN1A, FLT3, JUNB, MAP2K3, MYBPC2, RGS14, RNASEK, TBC1D30, and TXNDC11) derived from the PPI network of 336 genes were upregulated and may be associated with a high risk of TMZ resistance; these genes were identified in both the TCGA and local cohorts. Furthermore, the expression patterns of ARID5A, CDKN1A, and MAP2K3 were identical to the gene signatures of TMZ-resistant cell lines. The identified enrichment levels of the two gene sets expressed in tumor and peritumoral regions are potentially helpful for evaluating TMZ resistance in GBM. Moreover, these key genes could be used as biomarkers, potentially providing new molecular strategies for GBM treatment.

Detection of PD-L1 Expression in Temozolomide-Resistant Glioblastoma by Using PD-L1 Antibodies Conjugated with Lipid­Coated Superparamagnetic Iron Oxide.

PURPOSE: Targeted superparamagnetic iron oxide (SPIO) nanoparticles are a promising tool for molecular magnetic resonance imaging (MRI) diagnosis. Lipid-coated SPIO nanoparticles have a nonfouling property that can reduce nonspecific binding to off-target cells and prevent agglomeration, making them suitable contrast agents for molecular MRI diagnosis. PD-L1 is a poor prognostic factor for patients with glioblastoma. Most recurrent glioblastomas are temozolomide resistant. Diagnostic probes targeting PD-L1 could facilitate early diagnosis and be used to predict responses to targeted PD-L1 immunotherapy in patients with primary or recurrent glioblastoma. We conjugated lipid-coated SPIO nanoparticles with PD-L1 antibodies to identify PD-L1 expression in glioblastoma or temozolomide-resistant glioblastoma by using MRI. METHODS: The synthesized PD-L1 antibody-conjugated SPIO (PDL1-SPIO) nanoparticles were characterized using dynamic light scattering, zeta potential assays, transmission electron microscopy images, Prussian blue assay, in vitro cell affinity assay, and animal MRI analysis. RESULTS: PDL1-SPIO exhibited a specific binding capacity to PD-L1 of the mouse glioblastoma cell line (GL261). The presence and quantity of PDL1-SPIO in temozolomide-resistant glioblastoma cells and tumor tissue were confirmed through Prussian blue staining and in vivo T2* map MRI, respectively. CONCLUSION: This is the first study to demonstrate that PDL1-SPIO can specifically target temozolomide-resistant glioblastoma with PD-L1 expression in the brain and can be quantified through MRI analysis, thus making it suitable for the diagnosis of PD-L1 expression in temozolomide-resistant glioblastoma in vivo.

Diffusion Tensor Imaging of the Spinal Cord.

Spinal cord often is regarded as one of the last territories in the central nervous system where diffusion tensor imaging (DTI) can be used to probe white matter architecture. This article reviews current progress in spinal cord DTI, starting with anatomic properties and technical challenges that make spinal cord DTI a difficult task. Several possibilities offered by advanced pulse sequences that might overcome the difficulties are addressed, with associated trade-offs and limitations. Potential clinical assistance also is discussed in various spinal cord pathologies, such as myelopathy due to external compression, spinal cord tumors, acute ischemia, traumatic injury, and so forth.

Impact of physiological noise in characterizing the functional MRI default-mode network in Alzheimer's disease.

The functional connectivity of the default-mode network (DMN) monitored by functional magnetic resonance imaging (fMRI) in Alzheimer's disease (AD) patients has been found weaker than that in healthy participants. Since breathing and heart beating can cause fluctuations in the fMRI signal, these physiological activities may affect the fMRI data differently between AD patients and healthy participants. We collected resting-state fMRI data from AD patients and age-matched healthy participants. With concurrent cardiac and respiratory recordings, we estimated both physiological responses phase-locked and non-phase-locked to heart beating and breathing. We found that the cardiac and respiratory physiological responses in AD patients were 3.00 ± 0.51 s and 3.96 ± 0.52 s later (both p < 0.0001) than those in healthy participants, respectively. After correcting the physiological noise in the resting-state fMRI data by population-specific physiological response functions, the DMN estimated by seed-correlation was more localized to the seed region. The DMN difference between AD patients and healthy controls became insignificant after suppressing physiological noise. Our results indicate the importance of controlling physiological noise in the resting-state fMRI analysis to obtain clinically related characterizations in AD.

Radiomic Immunophenotyping of GSEA-Assessed Immunophenotypes of Glioblastoma and Its Implications for Prognosis: A Feasibility Study.

Characterization of immunophenotypes in glioblastoma (GBM) is important for therapeutic stratification and helps predict treatment response and prognosis. Radiomics can be used to predict molecular subtypes and gene expression levels. However, whether radiomics aids immunophenotyping prediction is still unknown. In this study, to classify immunophenotypes in patients with GBM, we developed machine learning-based magnetic resonance (MR) radiomic models to evaluate the enrichment levels of four immune subsets: Cytotoxic T lymphocytes (CTLs), activated dendritic cells, regulatory T cells (Tregs), and myeloid-derived suppressor cells (MDSCs). Independent testing data and the leave-one-out cross-validation method were used to evaluate model effectiveness and model performance, respectively. We identified five immunophenotypes (G1 to G5) based on the enrichment level for the four immune subsets. G2 had the worst prognosis and comprised highly enriched MDSCs and lowly enriched CTLs. G3 had the best prognosis and comprised lowly enriched MDSCs and Tregs and highly enriched CTLs. The average accuracy of T1-weighted contrasted MR radiomics models of the enrichment level for the four immune subsets reached 79% and predicted G2, G3, and the "immune-cold" phenotype (G1) according to our radiomics models. Our radiomic immunophenotyping models feasibly characterize the immunophenotypes of GBM and can predict patient prognosis.

Prediction of Recurrence-associated Death from Localized Prostate Cancer with a Charlson Comorbidity Index-reinforced Machine Learning Model.

Research has failed to resolve the dilemma experienced by localized prostate cancer patients who must choose between radical prostatectomy (RP) and external beam radiotherapy (RT). Because the Charlson Comorbidity Index (CCI) is a measurable factor that affects survival events, this research seeks to validate the potential of the CCI to improve the accuracy of various prediction models. Thus, we employed the Cox proportional hazard model and machine learning methods, including random forest (RF) and support vector machine (SVM), to model the data of medical records in the National Health Insurance Research Database (NHIRD). In total, 8581 individuals were enrolled, of whom 4879 had received RP and 3702 had received RT. Patients in the RT group were older and exhibited higher CCI scores and higher incidences of some CCI items. Moderate-to-severe liver disease, dementia, congestive heart failure, chronic pulmonary disease, and cerebrovascular disease all increase the risk of overall death in the Cox hazard model. The CCI-reinforced SVM and RF models are 85.18% and 81.76% accurate, respectively, whereas the SVM and RF models without the use of the CCI are relatively less accurate, at 75.81% and 74.83%, respectively. Therefore, CCI and some of its items are useful predictors of overall and prostate-cancer-specific survival and could constitute valuable features for machine-learning modeling.

Relative latency and temporal variability of hemodynamic responses at the human primary visual cortex.

The blood-oxygen-level-dependent (BOLD) functional MRI (fMRI) signal is a robust surrogate for local neuronal activity. However, it has been shown to vary substantially across subjects, brain regions, and repetitive measurements. This variability represents a limit to the precision of the BOLD response and the ability to reliably discriminate brain hemodynamic responses elicited by external stimuli or behavior that are nearby in time. While the temporal variability of the BOLD signal at human visual cortex has been found in the range of a few hundreds of milliseconds, the spatial distributions of the average and standard deviation of this temporal variability have not been quantitatively characterized. Here we use fMRI measurements with a high sampling rate (10Hz) to map the latency, intra- and inter-subject variability of the evoked BOLD signal in human primary (V1) visual cortices using an event-related fMRI paradigm. The latency relative to the average BOLD signal evoked by 30 stimuli was estimated to be 0.03±0.20s. Within V1, the absolute value of the relative BOLD latency was found correlated to intra- and inter-subject temporal variability. After comparing these measures to retinotopic maps, we found that locations with V1 areas sensitive to smaller eccentricity have later responses and smaller inter-subject variabilities. These correlations were found from data with either short inter-stimulus interval (ISI; average 4s) or long ISI (average 30s). Maps of the relative latency as well as inter-/intra-subject variability were found visually asymmetric between hemispheres. Our results suggest that the latency and variability of regional BOLD signal measured with high spatiotemporal resolution may be used to detect regional differences in hemodynamics to inform fMRI studies. However, the physiological origins of timing index distributions and their hemispheric asymmetry remain to be investigated.