Extra-Basal Ganglia Brain Structures Are Related to Motor Reserve in Parkinson's Disease

BACKGROUND

The "motor reserve" is an emerging concept based on the discrepancy between the severity of parkinsonism and dopaminergic degeneration; however, the related brain structures have not yet been elucidated.

OBJECTIVE

We investigated brain structures relevant to the motor reserve in Parkinson's disease (PD) in this study.

METHODS

Patients with drug-naïve, early PD were enrolled, who then underwent dopamine transporter (DAT) scan and diffusion tensor imaging (DTI). The severity of motor symptoms was evaluated with the Unified Parkinson's Disease Rating Scale score of bradykinesia and rigidity on the more affected side and dopaminergic degeneration of DAT uptake of the more affected putamen. Individual motor reserve estimate (MRE) was evaluated based on the discrepancy between the severity of motor symptoms and dopaminergic degeneration. Using DTI and the Brainnetome atlas, brain structures correlated with MRE were identified.

RESULTS

We enrolled 193 patients with drug-naïve PD (mean disease duration of 15.6±13.2 months), and the MRE successfully predicted the increase of levodopa equivalent dose after two years. In the DTI analysis, fractional anisotropy values of medial, inferior frontal, and temporal lobes, limbic structures, nucleus accumbens, and thalamus were positively correlated with the MRE, while no brain structures were correlated with mean diffusivity. Additionally, degree centrality derived from the structural connectivity of the frontal and temporal lobes and limbic structures was positively correlated with the MRE.

CONCLUSION

Our results show empirical evidence for MR in PD and brain structures relevant to MR, particularly, the extra-basal ganglia system including the limbic and frontal structures.

Tumor-Attentive Segmentation-Guided GAN for Synthesizing Breast Contrast-Enhanced MRI Without Contrast Agents

OBJECTIVE

Breast dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) is a sensitive imaging technique critical for breast cancer diagnosis. However, the administration of contrast agents poses a potential risk. This can be avoided if contrast-enhanced MRI can be obtained without using contrast agents. Thus, we aimed to generate T1-weighted contrast-enhanced MRI (ceT1) images from pre-contrast T1 weighted MRI (preT1) images in the breast.

METHODS

We proposed a generative adversarial network to synthesize ceT1 from preT1 breast images that adopted a local discriminator and segmentation task network to focus specifically on the tumor region in addition to the whole breast. The segmentation network performed a related task of segmentation of the tumor region, which allowed important tumor-related information to be enhanced. In addition, edge maps were included to provide explicit shape and structural information. Our approach was evaluated and compared with other methods in the local (n = 306) and external validation (n = 140) cohorts. Four evaluation metrics of normalized mean squared error (NRMSE), Pearson cross-correlation coefficients (CC), peak signal-to-noise ratio (PSNR), and structural similarity index map (SSIM) for the whole breast and tumor region were measured. An ablation study was performed to evaluate the incremental benefits of various components in our approach.

RESULTS

Our approach performed the best with an NRMSE 25.65, PSNR 54.80 dB, SSIM 0.91, and CC 0.88 on average, in the local test set.

CONCLUSION

Performance gains were replicated in the validation cohort.

SIGNIFICANCE

We hope that our method will help patients avoid potentially harmful contrast agents. Clinical and Translational Impact Statement-Contrast agents are necessary to obtain DCE-MRI which is essential in breast cancer diagnosis. However, administration of contrast agents may cause side effects such as nephrogenic systemic fibrosis and risk of toxic residue deposits. Our approach can generate DCE-MRI without contrast agents using a generative deep neural network. Thus, our approach could help patients avoid potentially harmful contrast agents resulting in an improved diagnosis and treatment workflow for breast cancer.

Robust multimodal fusion network using adversarial learning for brain tumor grading

BACKGROUND AND OBJECTIVE

Gliomas are graded using multimodal magnetic resonance imaging, which provides important information for treatment and prognosis. When modalities are missing, the grading is degraded. We propose a robust brain tumor grading model that can handle missing modalities.

METHODS

Our method was developed and tested on Brain Tumor Segmentation Challenge 2017 dataset (n = 285) via nested five-fold cross-validation. Our method adopts adversarial learning to generate the features of missing modalities relative to the features obtained from a full set of modalities in the latent space. An attention-based fusion block across modalities fuses the features of each available modality into a shared representation. Our method's results are compared to those of two other models where 15 missing-modality scenarios are explicitly considered and a joint training approach with random dropouts is used.

RESULTS

Our method outperforms the two competing methods in classifying high-grade gliomas (HGGs) and low-grade gliomas (LGGs), achieving an area under the curve of 87.76% on average for all missing-modality scenarios. The activation maps derived with our method confirm that it focuses on the enhancing portion of the tumor in HGGs and on the edema and non-enhancing portions of the tumor in LGGs, which is consistent with prior expertise. An ablation study shows the added benefits of a fusion block and adversarial learning for handling missing modalities.

CONCLUSION

Our method shows robust grading of gliomas in all cases of missing modalities. Our proposed network might have positive implications in glioma care by learning features robust to missing modalities.

Deep learning analysis to predict EGFR mutation status in lung adenocarcinoma manifesting as pure ground-glass opacity nodules on CT

Background

Epidermal growth factor receptor-tyrosine kinase inhibitors (EGFR-TKIs) showed potency as a non-invasive therapeutic approach in pure ground-glass opacity nodule (pGGN) lung adenocarcinoma. However, optimal methods of extracting information about EGFR mutation from pGGN lung adenocarcinoma images remain uncertain. We aimed to develop, validate, and evaluate the clinical utility of a deep learning model for predicting EGFR mutation status in lung adenocarcinoma manifesting as pGGN on computed tomography (CT).

Methods

We included 185 resected pGGN lung adenocarcinomas in the primary cohort. The patients were divided into training (n = 125), validation (n = 23), and test sets (n = 37). A preoperative CT-based deep learning model with clinical factors as well as clinical and radiomics models was constructed and applied to the test set. We evaluated the clinical utility of the deep learning model by applying it to 83 GGNs that received EGFR-TKI from an independent cohort (clinical validation set), and treatment response was regarded as the reference standard.

Results

The prediction efficiencies of each model were compared in terms of area under the curve (AUC). Among the 185 pGGN lung adenocarcinomas, 122 (65.9%) were EGFR-mutant and 63 (34.1%) were EGFR-wild type. The AUC of the clinical, radiomics, and deep learning with clinical models to predict EGFR mutations were 0.50, 0.64, and 0.85, respectively, for the test set. The AUC of deep learning with the clinical model in the validation set was 0.72.

Conclusions

Deep learning approach of CT images combined with clinical factors can predict EGFR mutations in patients with lung adenocarcinomas manifesting as pGGN, and its clinical utility was demonstrated in a real-world sample.

Best cutoff score of cervical-pedicle thickness as a morphological parameter for predicting cervical central stenosis

There are various factors for the cause of cervical central stenosis (CCS), such as osteophyte, cervical-disc degeneration, and cervical ligamentum flavum hypertrophy. However, the pedicle of the cervical vertebra has not yet been analyzed for its relationship with CCS. We created a new morphologic parameter called the cervical-pedicle thickness (CPT) to assess the association between CCS and the cervical pedicle. We obtained morphological cases involving the CPT from 82 patients with CCS. There were also 84 in the normal group who underwent cervical spine magnetic resonance imaging (CS-MR) as part of routine health screening. We obtained the T2-weighted CS-MR axial images from group members, and assessed the CPT at the level of the C6 vertebra on CS-MR. The mean CPT was 3.46 ± 0.57 mm in the normal group, 4.97 ± 0.75 mm in the CCS group, which thus had a significantly higher CPT (P < .01) than did the normal group. For the prognostic value of the CPT as a predictor of CCS, ROC analysis indicated that the best cutoff score for the CPT was 4.18 mm, with 93.9% sensitivity, 92.9% specificity, and AUC 0.97. Greater CPT was highly associated with a possibility of CCS. This conclusion will be helpful for assessing the CCS patients.

Diagnosis-informed connectivity subtyping discovers subgroups of autism with reproducible symptom profiles

Clinical heterogeneity has been one of the main barriers to develop effective biomarkers and therapeutic strategies in autism spectrum disorder (ASD). Recognizing this challenge, much effort has been made in recent neuroimaging studies to find biologically more homogeneous subgroups (called 'neurosubtypes') in autism. However, most approaches have rarely evaluated how much the employed features in subtyping represent the core anomalies of ASD, obscuring its utility in actual clinical diagnosis. To address this, we combined two data-driven methods, 'connectome-based gradient' and 'functional random forest', collectively allowing to discover reproducible neurosubtypes based on resting-state functional connectivity profiles that are specific to ASD. Indeed, the former technique provides the features (as input for subtyping) that effectively summarize whole-brain connectome variations in both normal and ASD conditions, while the latter leverages a supervised random forest algorithm to inform diagnostic labels to clustering, which makes neurosubtyping driven by the features of ASD core anomalies. Applying this framework to the open-sharing Autism Brain Imaging Data Exchange repository data (discovery, n = 103/108 for ASD/typically developing [TD]; replication, n = 44/42 for ASD/TD), we found three dominant subtypes of functional gradients in ASD and three subtypes in TD. The subtypes in ASD revealed distinct connectome profiles in multiple brain areas, which are associated with different Neurosynth-derived cognitive functions previously implicated in autism studies. Moreover, these subtypes showed different symptom severity, which degree co-varies with the extent of functional gradient changes observed across the groups. The subtypes in the discovery and replication datasets showed similar symptom profiles in social interaction and communication domains, confirming a largely reproducible brain-behavior relationship. Finally, the connectome gradients in ASD subtypes present both common and distinct patterns compared to those in TD, reflecting their potential overlap and divergence in terms of developmental mechanisms involved in the manifestation of large-scale functional networks. Our study demonstrated a potential of the diagnosis-informed subtyping approach in developing a clinically useful brain-based classification system for future ASD research.

Measurement of Perfusion Heterogeneity within Tumor Habitats on Magnetic Resonance Imaging and Its Association with Prognosis in Breast Cancer Patients

Simple Summary

A habitat analysis reflects intratumoral heterogeneity more accurately than does a whole-tumor analysis. Perfusional heterogeneity using a habitat analysis is a rarely explored option and can affect patient outcomes. From two hospitals, 308 and 147 patients with invasive breast cancer who underwent preoperative MRI were included as development and validation cohorts, respectively. In our study, five habitats with distinct perfusion patterns were identified based on early and delayed phases of dynamic contrast material-enhanced MR images. A habitat risk score (HRS) was an independent risk factor for predicting worse disease-free survival outcomes in the HRS-only risk model (hazard ratio = 3.274 [95% CI = 1.378–7.782]; p = 0.014) and combined habitat risk model (hazard ratio = 4.128 [95% CI = 1.744–9.769]; p = 0.003) in the validation cohort.

Abstract

The purpose of this study was to identify perfusional subregions sharing similar kinetic characteristics from dynamic contrast-enhanced magnetic resonance imaging (MRI) using data-driven clustering, and to evaluate the effect of perfusional heterogeneity based on those subregions on patients’ survival outcomes in various risk models. From two hospitals, 308 and 147 women with invasive breast cancer who underwent preoperative MRI between October 2011 and July 2012 were retrospectively enrolled as development and validation cohorts, respectively. Using the Cox-least absolute shrinkage and selection operator model, a habitat risk score (HRS) was constructed from the radiomics features from the derived habitat map. An HRS-only, clinical, combined habitat, and two conventional radiomics risk models to predict patients’ disease-free survival (DFS) were built. Patients were classified into low-risk or high-risk groups using the median cutoff values of each risk score. Five habitats with distinct perfusion patterns were identified. An HRS was an independent risk factor for predicting worse DFS outcomes in the HRS-only risk model (hazard ratio = 3.274 [95% CI = 1.378–7.782]; p = 0.014) and combined habitat risk model (hazard ratio = 4.128 [95% CI = 1.744–9.769]; p = 0.003) in the validation cohort. In the validation cohort, the combined habitat risk model (hazard ratio = 4.128, p = 0.003, C-index = 0.760) showed the best performance among five different risk models. The quantification of perfusion heterogeneity is a potential approach for predicting prognosis and may facilitate personalized, tailored treatment strategies for breast cancer.

A Cascaded Neural Network for Staging in Non-Small Cell Lung Cancer Using Pre-Treatment CT

BACKGROUND AND AIM

Tumor staging in non-small cell lung cancer (NSCLC) is important for treatment and prognosis. Staging involves expert interpretation of imaging, which we aim to automate with deep learning (DL). We proposed a cascaded DL method comprised of two steps to classification between early- and advanced-stage NSCLC using pretreatment computed tomography.

METHODS

We developed and tested a DL model to classify between early- and advanced-stage using training (n = 90), validation (n = 8), and two test (n = 37, n = 26) cohorts obtained from the public domain. The first step adopted an autoencoder network to compress the imaging data into latent variables and the second step used the latent variable to classify the stages using the convolutional neural network (CNN). Other DL and machine learning-based approaches were compared.

RESULTS

Our model was tested in two test cohorts of CPTAC and TCGA. In CPTAC, our model achieved accuracy of 0.8649, sensitivity of 0.8000, specificity of 0.9412, and area under the curve (AUC) of 0.8206 compared to other approaches (AUC 0.6824-0.7206) for classifying between early- and advanced-stages. In TCGA, our model achieved accuracy of 0.8077, sensitivity of 0.7692, specificity of 0.8462, and AUC of 0.8343.

CONCLUSION

Our cascaded DL model for classification NSCLC patients into early-stage and advanced-stage showed promising results and could help future NSCLC research.

Pleomorphic carcinoma of the lung: Prognostic models of semantic, radiomics and combined features from CT and PET/CT in 85 patients

Introduction

To demonstrate semantic, radiomics, and the combined risk models related to the prognoses of pulmonary pleomorphic carcinomas (PCs).

Methods

We included 85 patients (M:F = 71:14; age, 35–88 [mean, 63 years]) whose imaging features were divided into training (n = 60) and test (n = 25) sets. Nineteen semantic and 142 radiomics features related to tumors were computed. Semantic risk score (SRS) model was built using the Cox-least absolute shrinkage and selection operator (LASSO) approach. Radiomics risk score (RRS) from CT and PET features and combined risk score (CRS) adopting both semantic and radiomics features were also constructed. Risk groups were stratified by the median of the risk scores of the training set. Survival analysis was conducted with the Kaplan-Meier plots.

Results

Of 85 PCs, adenocarcinoma was the most common epithelial component found in 63 (73 %) tumors. In SRS model, four features were stratified into high- and low-risk groups (HR, 4.119; concordance index ([C-index], 0.664) in the test set. In RRS model, five features helped improve the stratification (HR, 3.716; C-index, 0.591) and in CRS model, three features helped perform the best stratification (HR, 4.795; C-index, 0.617). The two significant features of CRS models were the SUVmax and the histogram feature of energy ([CT Firstorder Energy]).

Conclusion

In PCs of the lungs, the combined model leveraging semantic and radiomics features provides a better prognosis compared to using semantic and radiomics features separately. The high SUVmax of solid portion (CT Firstorder Energy) of tumors is associated with poor prognosis in lung PCs.

Accurate neuroimaging biomarkers to predict body mass index in adolescents: a longitudinal study

Obesity is often associated with cardiovascular complications. Adolescent obesity is a risk factor for cardiovascular disease in adulthood; thus, intensive management is warranted in adolescence. The brain state contributes to the development of obesity in addition to metabolic conditions, and hence neuroimaging is an important tool for accurately assessing an individual's risk of developing obesity. Here, we aimed to predict body mass index (BMI) progression in adolescents with neuroimaging features using machine learning approaches. From an open database, we adopted 76 resting-state functional magnetic resonance imaging (rs-fMRI) datasets from adolescents with longitudinal BMI scores. Functional connectivity analyses were performed on cortical surfaces and subcortical volumes. We identified baseline functional connectivity features in the prefrontal-, posterior cingulate-, sensorimotor-, and inferior parietal-cortices as significant determinants of BMI changes. A BMI prediction model based on the identified fMRI biomarkers exhibited a high accuracy (intra-class correlation = 0.98) in predicting BMI at the second visit (1~2 years later). The identified brain regions were significantly correlated with the eating disorder-, anxiety-, and depression-related scores. Based on these results, we concluded that these functional connectivity features in brain regions related to eating disorders and emotional processing could be important neuroimaging biomarkers for predicting BMI progression.

Are radiomics features universally applicable to different organs?

BACKGROUND

Many studies have successfully identified radiomics features reflecting macroscale tumor features and tumor microenvironment for various organs. There is an increased interest in applying these radiomics features found in a given organ to other organs. Here, we explored whether common radiomics features could be identified over target organs in vastly different environments.

METHODS

Four datasets of three organs were analyzed. One radiomics model was constructed from the training set (lungs, n = 401), and was further evaluated in three independent test sets spanning three organs (lungs, n = 59; kidneys, n = 48; and brains, n = 43). Intensity histograms derived from the whole organ were compared to establish organ-level differences. We constructed a radiomics score based on selected features using training lung data over the tumor region. A total of 143 features were computed for each tumor. We adopted a feature selection approach that favored stable features, which can also capture survival. The radiomics score was applied to three independent test data from lung, kidney, and brain tumors, and whether the score could be used to separate high- and low-risk groups, was evaluated.

RESULTS

Each organ showed a distinct pattern in the histogram and the derived parameters (mean and median) at the organ-level. The radiomics score trained from the lung data of the tumor region included seven features, and the score was only effective in stratifying survival for other lung data, not in other organs such as the kidney and brain. Eliminating the lung-specific feature (2.5 percentile) from the radiomics score led to similar results. There were no common features between training and test sets, but a common category of features (texture category) was identified.

CONCLUSION

Although the possibility of a generally applicable model cannot be excluded, we suggest that radiomics score models for survival were mostly specific for a given organ; applying them to other organs would require careful consideration of organ-specific properties.

Tumor Margin Contains Prognostic Information: Radiomic Margin Characteristics Analysis in Lung Adenocarcinoma Patients

Simple Summary

The tumor microenvironment is a dynamic area, with continuous interaction between tumor cells and their surrounding environment. We aimed to investigate the relationship between tumor radiomic margin characteristics and prognosis in patients with lung cancer. When compared to the model with clinical variables only (C-index = 0.738), the model incorporating clinical variables and radiomic margin characteristics (C-index = 0.753) demonstrated a higher C-index for predicting overall survival. In the model integrating both clinical variables and radiomic margin characteristics, convexity, Laplace of Gaussian (LoG) kurtosis 3, and roundness factor were independent predictive factors of overall survival. Our study showed that radiomic margin characteristics helped predict overall survival in patients with lung adenocarcinomas, thus implying that the tumor margin contains prognostic information.

Abstract

We aimed to investigate the relationship between tumor radiomic margin characteristics and prognosis in patients with lung cancer. We enrolled 334 patients who underwent complete resection for lung adenocarcinoma. A quantitative computed tomography analysis was performed, and 76 radiomic margin characteristics were extracted. The radiomic margin characteristics were correlated with overall survival. The selected clinical variables and radiomic margin characteristics were used to calculate a prognostic model with subsequent internal and external validation. Nearly all of the radiomic margin characteristics showed excellent reproducibility. The least absolute shrinkage and selection operator (LASSO) method was used to select eight radiomic margin characteristics. When compared to the model with clinical variables only (C-index = 0.738), the model incorporating clinical variables and radiomic margin characteristics (C-index = 0.753) demonstrated a higher C-index for predicting overall survival. In the model integrating both clinical variables and radiomic margin characteristics, convexity, a Laplace of Gaussian (LoG) kurtosis of 3, and the roundness factor were each independently predictive of overall survival. In addition, radiomic margin characteristics were also correlated with the micropapillary subtype, and the sphericity value was able to predict the presence of the micropapillary subtype. In conclusion, our study showed that radiomic margin characteristics helped predict overall survival in patients with lung adenocarcinomas, thus implying that the tumor margin contains prognostic information.

Predicting amyloid positivity in patients with mild cognitive impairment using a radiomics approach

Predicting amyloid positivity in patients with mild cognitive impairment (MCI) is crucial. In the present study, we predicted amyloid positivity with structural MRI using a radiomics approach. From MR images (including T1, T2 FLAIR, and DTI sequences) of 440 MCI patients, we extracted radiomics features composed of histogram and texture features. These features were used alone or in combination with baseline non-imaging predictors such as age, sex, and ApoE genotype to predict amyloid positivity. We used a regularized regression method for feature selection and prediction. The performance of the baseline non-imaging model was at a fair level (AUC = 0.71). Among single MR-sequence models, T1 and T2 FLAIR radiomics models also showed fair performances (AUC for test = 0.71-0.74, AUC for validation = 0.68-0.70) in predicting amyloid positivity. When T1 and T2 FLAIR radiomics features were combined, the AUC for test was 0.75 and AUC for validation was 0.72 (p vs. baseline model < 0.001). The model performed best when baseline features were combined with a T1 and T2 FLAIR radiomics model (AUC for test = 0.79, AUC for validation = 0.76), which was significantly better than those of the baseline model (p < 0.001) and the T1 + T2 FLAIR radiomics model (p < 0.001). In conclusion, radiomics features showed predictive value for amyloid positivity. It can be used in combination with other predictive features and possibly improve the prediction performance.

Computed Tomography Radiomics for Residual Positron Emission Tomography-Computed Tomography Uptake in Lymph Nodes after Treatment

Simple Summary

In this study we explored the predictive ability of radiomics in non-small cell lung cancer patients, and reported the complementary role of radiomics in predicting the treatment response of the lymph nodes. Radiomics analysis is a cutting-edge technology for the noninvasive assessment of tumor biology, which converts medical images into mineable high-dimensional data. Our method is cost-effective with no need for additional studies, and moreover, we used an easily reproducible study method that can be applicable in further studies using radiomics in oncology.

Abstract

Although a substantial decrease in 2-[fluorine-18]fluoro-2-deoxy-d-glucose (FDG) uptake on positron emission tomography-computed tomography (PET-CT) indicates a promising metabolic response to treatment, predicting the pathologic status of lymph nodes (LN) remains challenging. We investigated the potential of a CT radiomics approach to predict the pathologic complete response of LNs showing residual uptake after neoadjuvant concurrent chemoradiotherapy (NeoCCRT) in patients with non-small cell lung cancer (NSCLC). Two hundred and thirty-seven patients who underwent NeoCCRT for stage IIIa NSCLC were included. Two hundred fifty-two CT radiomics features were extracted from LNs showing remaining positive FDG uptake upon restaging PET-CT. A multivariable logistic regression analysis of radiomics features and clinicopathologic characteristics was used to develop a prediction model. Of the 237 patients, 135 patients (185 nodes) met our inclusion criteria. Eighty-seven LNs were proven to be malignant (47.0%, 87/185). Upon multivariable analysis, metastatic LNs were significantly prevalent in females and patients with adenocarcinoma (odds ratio (OR) = 2.02, 95% confidence interval (CI) = 0.88–4.62 and OR = 0.39, 95% CI = 0.19–0.77 each). Metastatic LNs also had a larger maximal 3D diameter and higher cluster tendency (OR = 9.92, 95% CI = 3.15–31.17 and OR = 2.36, 95% CI = 1.22–4.55 each). The predictive model for metastasis showed a discrimination performance with an area under the receiver operating characteristic curve of 0.728 (95% CI = 0.654–0.801, p value < 0.001). The radiomics approach allows for the noninvasive detection of metastases in LNs with residual FDG uptake after the treatment of NSCLC patients.

The Tumor-Fat Interface Volume of Breast Cancer on Pretreatment MRI Is Associated with a Pathologic Response to Neoadjuvant Chemotherapy

Simple Summary

Contact between a tumor and the adjacent fat is a potential biomarker to predict the therapy response in breast cancer, but it has not been quantitatively explored. In this study, we measured the direct contact between the tumor and adjacent fat using breast magnetic resonance imaging with machine learning and found that patients with a greater volume of contact between tumor and fat were less likely to have a complete pathological response. Our results suggest that the volume of the tumor–fat interface is a potential prognostic imaging biomarker to predict the treatment response to neoadjuvant chemotherapy.

Abstract

Adipocytes are active sources of numerous adipokines that work in both a paracrine and endocrine manner. It is not known that the direct contact between tumor and neighboring fat measured by pretreatment breast magnetic resonance imaging (MRI) affects treatment outcomes to neoadjuvant chemotherapy (NAC) in breast cancer patients. A biomarker quantifying the tumor–fat interface volume from pretreatment MRI was proposed and used to predict pathologic complete response (pCR) in breast cancer patients treated with NAC. The tumor–fat interface volume was computed with data-driven clustering using multiphasic MRI. Our approach was developed and validated in two cohorts consisting of 1140 patients. A high tumor–fat interface volume was significantly associated with a non-pCR in both the development and validation cohorts (p = 0.030 and p = 0.037, respectively). Quantitative measurement of the tumor–fat interface volume based on pretreatment MRI may be useful for precision medicine and subsequently influence the treatment strategy of patients.

GraphNet-based imaging biomarker model to explain levodopa-induced dyskinesia in Parkinson's disease

BACKGROUND AND OBJECTIVE

Levodopa-induced dyskinesia (LID) is a disabling complication of Parkinson's disease (PD). Imaging-based measurements, especially those related to the surface shape of the basal ganglia, have shown potential for explaining the severity of LID in PD. Here, we aimed to explore a novel application of the methodology to find biomarkers of LID severity in PD using regularization.

METHODS

We proposed an application of graph-constrained elastic net (GraphNet) regularization to detect surface-based shape biomarkers explaining the severity of LID and compared the approach with other conventional regularization methods. To examine the methods, we used two independent datasets, one as a training dataset to build the model, and the other dataset was used to validate the constructed model.

RESULTS

We found that the left striatum (putamen was the greatest and the caudate was second) was the most significant surface-based biomarker related to the severity of LID. Our results improved the interpretability of identified surface-based biomarkers compared to competing methods. We also found that GraphNet regularization improved prediction of the severity of LID better than the conventional regularization methods. Our model performed better in terms of root-mean-squared error and correlation coefficient between predicted and actual clinical scores.

CONCLUSION

The proposed algorithm offers an advantage of interpretable anatomical variations related to the deformation of the cortical surface. The experimental results showed that GraphNet regularization was robust to identify surface-based shape biomarkers related to both hypokinetic and hyperkinetic movement disorders.

A neuroimaging marker for predicting longitudinal changes in pain intensity of subacute back pain based on large-scale brain network interactions

Identification of predictive neuroimaging markers of pain intensity changes is a crucial issue to better understand macroscopic neural mechanisms of pain. Although a single connection between the medial prefrontal cortex and nucleus accumbens has been suggested as a powerful marker, how the complex interactions on a large-scale brain network can serve as the markers is underexplored. Here, we aimed to identify a set of functional connections predictive of longitudinal changes in pain intensity using large-scale brain networks. We re-analyzed previously published resting-state functional magnetic resonance imaging data of 49 subacute back pain (SBP) patients. We built a network-level model that predicts changes in pain intensity over one year by combining independent component analysis and a penalized regression framework. Connections involving top-down pain modulation, multisensory integration, and mesocorticolimbic circuits were identified as predictive markers for pain intensity changes. Pearson’s correlations between actual and predicted pain scores were r = 0.33–0.72, and group classification results between SBP patients with persisting pain and recovering patients, in terms of area under the curve (AUC), were 0.89/0.75/0.75 for visits four/three/two, thus outperforming the previous work (AUC 0.83/0.73/0.67). This study identified functional connections important for longitudinal changes in pain intensity in SBP patients, providing provisional markers to predict future pain using large-scale brain networks.

Machine learning-based automated classification of headache disorders using patient-reported questionnaires

Classification of headache disorders is dependent on a subjective self-report from patients and its interpretation by physicians. We aimed to apply objective data-driven machine learning approaches to analyze patient-reported symptoms and test the feasibility of the automated classification of headache disorders. The self-report data of 2162 patients were analyzed. Headache disorders were merged into five major entities. The patients were divided into training (n = 1286) and test (n = 876) cohorts. We trained a stacked classifier model with four layers of XGBoost classifiers. The first layer classified between migraine and others, the second layer classified between tension-type headache (TTH) and others, and the third layer classified between trigeminal autonomic cephalalgia (TAC) and others, and the fourth layer classified between epicranial and thunderclap headaches. Each layer selected different features from the self-reports by using least absolute shrinkage and selection operator. In the test cohort, our stacked classifier obtained accuracy of 81%, sensitivity of 88%, 69%, 65%, 53%, and 51%, and specificity of 95%, 55%, 46%, 48%, and 51% for migraine, TTH, TAC, epicranial headache, and thunderclap headaches, respectively. We showed that a machine-learning based approach is applicable in analyzing patient-reported questionnaires. Our result could serve as a baseline for future studies in headache research.

Whole-brain functional connectivity correlates of obesity phenotypes

Dysregulated neural mechanisms in reward and somatosensory circuits result in an increased appetitive drive for and reduced inhibitory control of eating, which in turn causes obesity. Despite many studies investigating the brain mechanisms of obesity, the role of macroscale whole‐brain functional connectivity remains poorly understood. Here, we identified a neuroimaging‐based functional connectivity pattern associated with obesity phenotypes by using functional connectivity analysis combined with machine learning in a large‐scale (n ~ 2,400) dataset spanning four independent cohorts. We found that brain regions containing the reward circuit positively associated with obesity phenotypes, while brain regions for sensory processing showed negative associations. Our study introduces a novel perspective for understanding how the whole‐brain functional connectivity correlates with obesity phenotypes. Furthermore, we demonstrated the generalizability of our findings by correlating the functional connectivity pattern with obesity phenotypes in three independent datasets containing subjects of multiple ages and ethnicities. Our findings suggest that obesity phenotypes can be understood in terms of macroscale whole‐brain functional connectivity and have important implications for the obesity neuroimaging community.

Multivariate association between brain function and eating disorders using sparse canonical correlation analysis

Eating disorder is highly associated with obesity and it is related to brain dysfunction as well. Still, the functional substrates of the brain associated with behavioral traits of eating disorder are underexplored. Existing neuroimaging studies have explored the association between eating disorder and brain function without using all the information provided by the eating disorder related questionnaire but by adopting summary factors. Here, we aimed to investigate the multivariate association between brain function and eating disorder at fine-grained question-level information. Our study is a retrospective secondary analysis that re-analyzed resting-state functional magnetic resonance imaging of 284 participants from the enhanced Nathan Kline Institute-Rockland Sample database. Leveraging sparse canonical correlation analysis, we associated the functional connectivity of all brain regions and all questions in the eating disorder questionnaires. We found that executive- and inhibitory control-related frontoparietal networks showed positive associations with questions of restraint eating, while brain regions involved in the reward system showed negative associations. Notably, inhibitory control-related brain regions showed a positive association with the degree of obesity. Findings were well replicated in the independent validation dataset (n = 34). The results of this study might contribute to a better understanding of brain function with respect to eating disorder.

Prediction of age at onset in Parkinson's disease using objective specific neuroimaging genetics based on a sparse canonical correlation analysis

The age at onset (AAO) is an important determinant in Parkinson’s disease (PD). Neuroimaging genetics is suitable for studying AAO in PD as it jointly analyzes imaging and genetics. We aimed to identify features associated with AAO in PD by applying the objective-specific neuroimaging genetics approach and constructing an AAO prediction model. Our objective-specific neuroimaging genetics extended the sparse canonical correlation analysis by an additional data type related to the target task to investigate possible associations of the imaging–genetic, genetic–target, and imaging–target pairs simultaneously. The identified imaging, genetic, and combined features were used to construct analytical models to predict the AAO in a nested five-fold cross-validation. We compared our approach with those from two feature selection approaches where only associations of imaging–target and genetic–target were explored. Using only imaging features, AAO prediction was accurate in all methods. Using only genetic features, the results from other methods were worse or unstable compared to our model. Using both imaging and genetic features, our proposed model predicted the AAO well (r = 0.5486). Our findings could have significant impacts on the characterization of prodromal PD and contribute to diagnosing PD early because genetic features could be measured accurately from birth.

Radiomics Based on Thyroid Ultrasound Can Predict Distant Metastasis of Follicular Thyroid Carcinoma

We aimed to evaluate whether radiomics analysis based on gray-scale ultrasound (US) can predict distant metastasis of follicular thyroid cancer (FTC). We retrospectively included 35 consecutive FTCs with distant metastases and 134 FTCs without distant metastasis. We extracted a total of 60 radiomics features derived from the first order, shape, gray-level cooccurrence matrix, and gray-level size zone matrix features using US imaging. A radiomics signature was generated using the least absolute shrinkage and selection operator and was used to train a support vector machine (SVM) classifier in five-fold cross-validation. The SVM classifier showed an area under the curve (AUC) of 0.90 on average on the test folds. Age, size, widely invasive histology, extrathyroidal extension, lymph node metastases on pathology, nodule-in-nodule appearance, marked hypoechogenicity, and rim calcification on the US were significantly more frequent among FTCs with distant metastasis compared to those without metastasis (p < 0.05). Radiomics signature and widely invasive histology were significantly associated with distant metastasis on multivariate analysis (p < 0.01 and p = 0.003). The classifier using the results of the multivariate analysis showed an AUC of 0.93. The radiomics signature from thyroid ultrasound is an independent biomarker for noninvasively predicting distant metastasis of FTC.

Multi-Habitat Radiomics Unravels Distinct Phenotypic Subtypes of Glioblastoma with Clinical and Genomic Significance

We aimed to evaluate the potential of radiomics as an imaging biomarker for glioblastoma (GBM) patients and explore the molecular rationale behind radiomics using a radio-genomics approach. A total of 144 primary GBM patients were included in this study (training cohort). Using multi-parametric MR images, radiomics features were extracted from multi-habitats of the tumor. We applied Cox-LASSO algorithm to build a survival prediction model, which we validated using an independent validation cohort. GBM patients were consensus clustered to reveal inherent phenotypic subtypes. GBM patients were successfully stratified by the radiomics risk score, a weighted sum of radiomics features, corroborating the potential of radiomics as a prognostic biomarker. Using consensus clustering, we identified three distinct subtypes which significantly differed in the prognosis ("heterogenous enhancing", "rim-enhancing necrotic", and "cystic" subtypes). Transcriptomic traits enriched in individual subtypes were in accordance with imaging phenotypes summarized by radiomics. For example, rim-enhancing necrotic subtype was well described by radiomics profiling (T2 autocorrelation and flat shape) and highlighted by the inflammatory genomic signatures, which well correlated to its phenotypic peculiarity (necrosis). This study showed that imaging subtypes derived from radiomics successfully recapitulated the genomic underpinnings of GBMs and thereby confirmed the feasibility of radiomics as an imaging biomarker for GBM patients with comprehensible biologic annotation.

Integrative Radiogenomics Approach for Risk Assessment of Post-Operative Metastasis in Pathological T1 Renal Cell Carcinoma: A Pilot Retrospective Cohort Study

Despite the increasing incidence of pathological stage T1 renal cell carcinoma (pT1 RCC), postoperative distant metastases develop in many surgically treated patients, causing death in certain cases. Therefore, this study aimed to create a radiomics model using imaging features from multiphase computed tomography (CT) to more accurately predict the postoperative metastasis of pT1 RCC and further investigate the possible link between radiomics parameters and gene expression profiles generated by whole transcriptome sequencing (WTS). Four radiomic features, including the minimum value of a histogram feature from inner regions of interest (ROIs) (INNER_Min_hist), the histogram of the energy feature from outer ROIs (OUTER_Energy_Hist), the maximum probability of gray-level co-occurrence matrix (GLCM) feature from inner ROIs (INNER_MaxProb_GLCM), and the ratio of voxels under 80 Hounsfield units (Hus) in the nephrographic phase of postcontrast CT (Under80HURatio), were detected to predict the postsurgical metastasis of patients with pathological stage T1 RCC, and the clinical outcomes of patients could be successfully stratified based on their radiomic risk scores. Furthermore, we identified heterogenous-trait-associated gene signatures correlated with these four radiomic features, which captured clinically relevant molecular pathways, tumor immune microenvironment, and potential treatment strategies. Our results of accurate surrogates using radiogenomics could lead to additional benefit from adjuvant therapy or postsurgical metastases in pT1 RCC.

Radiomics Study of Thyroid Ultrasound for Predicting BRAF Mutation in Papillary Thyroid Carcinoma: Preliminary Results

BACKGROUND AND PURPOSE

It is not known how radiomics using ultrasound images contribute to the detection of BRAF mutation. This study aimed to evaluate whether a radiomics study of gray-scale ultrasound can predict the presence or absence of B-Raf proto-oncogene, serine/threonine kinase (BRAF) mutation in papillary thyroid cancer.

MATERIALS AND METHODS

The study retrospectively included 96 thyroid nodules that were surgically confirmed papillary thyroid cancers between January 2012 and June 2013. BRAF mutation was positive in 48 nodules and negative in 48 nodules. For analysis, ROIs from the nodules were demarcated manually on both longitudinal and transverse sonographic images. We extracted a total of 86 radiomics features derived from histogram parameters, gray-level co-occurrence matrix, intensity size zone matrix, and shape features. These features were used to build 3 different classifier models, including logistic regression, support vector machine, and random forest using 5-fold cross-validation. The performance including accuracy, sensitivity, specificity, positive predictive value, negative predictive value, and area under the receiver operating characteristic curve, of the different models was evaluated.

RESULTS

The incidence of high-suspicion nodules diagnosed on ultrasound was higher in the BRAF mutation-positive group than in the mutation-negative group (P  =  .004). The radiomics approach demonstrated that all classification models showed moderate performance for predicting the presence of BRAF mutation in papillary thyroid cancers with an area under the curve value of 0.651, accuracy of 64.3%, sensitivity of 66.8%, and specificity of 61.8%, on average, for the 3 models.

CONCLUSIONS

Radiomics study using thyroid sonography is limited in predicting the BRAF mutation status of papillary thyroid carcinoma. Further studies will be needed to validate our results using various diagnostic methods.