Generalized ComBat harmonization methods for radiomic features with multi-modal distributions and multiple batch effects

DeepComBat: A statistically motivated, hyperparameter-robust, deep learning approach to harmonization of neuroimaging data

Neuroimaging data acquired using multiple scanners or protocols are increasingly available. However, such data exhibit technical artifacts across batches which introduce confounding and decrease reproducibility. This is especially true when multi-batch data are analyzed using complex downstream models which are more likely to pick up on and implicitly incorporate batch-related information. Previously proposed image harmonization methods have sought to remove these batch effects; however, batch effects remain detectable in the data after applying these methods. We present DeepComBat, a deep learning harmonization method based on a conditional variational autoencoder and the ComBat method. DeepComBat combines the strengths of statistical and deep learning methods in order to account for the multivariate relationships between features while simultaneously relaxing strong assumptions made by previous deep learning harmonization methods. As a result, DeepComBat can perform multivariate harmonization while preserving data structure and avoiding the introduction of synthetic artifacts. We apply this method to cortical thickness measurements from a cognitive-aging cohort and show DeepComBat qualitatively and quantitatively outperforms existing methods in removing batch effects while preserving biological heterogeneity. Additionally, DeepComBat provides a new perspective for statistically motivated deep learning harmonization methods.

Multivariate testing and effect size measures for batch effect evaluation in radiomic features

While precision medicine applications of radiomics analysis are promising, differences in image acquisition can cause "batch effects" that reduce reproducibility and affect downstream predictive analyses. Harmonization methods such as ComBat have been developed to correct these effects, but evaluation methods for quantifying batch effects are inconsistent. In this study, we propose the use of the multivariate statistical test PERMANOVA and the Robust Effect Size Index (RESI) to better quantify and characterize batch effects in radiomics data. We evaluate these methods in both simulated and real radiomics features extracted from full-field digital mammography (FFDM) data. PERMANOVA demonstrated higher power than standard univariate statistical testing, and RESI was able to interpretably quantify the effect size of site at extremely large sample sizes. These methods show promise as more powerful and interpretable methods for the detection and quantification of batch effects in radiomics studies.

Independent validation of CT radiomics models in colorectal liver metastases: predicting local tumour progression after ablation

OBJECTIVES

Independent internal and external validation of three previously published CT-based radiomics models to predict local tumor progression (LTP) after thermal ablation of colorectal liver metastases (CRLM).

MATERIALS AND METHODS

Patients with CRLM treated with thermal ablation were collected from two institutions to collect a new independent internal and external validation cohort. Ablation zones (AZ) were delineated on portal venous phase CT 2-8 weeks post-ablation. Radiomics features were extracted from the AZ and a 10 mm peri-ablational rim (PAR) of liver parenchyma around the AZ. Three previously published prediction models (clinical, radiomics, combined) were tested without retraining. LTP was defined as new tumor foci appearing next to the AZ up to 24 months post-ablation.

RESULTS

The internal cohort included 39 patients with 68 CRLM and the external cohort 52 patients with 78 CRLM. 34/146 CRLM developed LTP after a median follow-up of 24 months (range 5-139). The median time to LTP was 8 months (range 2-22). The combined clinical-radiomics model yielded a c-statistic of 0.47 (95%CI 0.30-0.64) in the internal cohort and 0.50 (95%CI 0.38-0.62) in the external cohort, compared to 0.78 (95%CI 0.65-0.87) in the previously published original cohort. The radiomics model yielded c-statistics of 0.46 (95%CI 0.29-0.63) and 0.39 (95%CI 0.28-0.52), and the clinical model 0.51 (95%CI 0.34-0.68) and 0.51 (95%CI 0.39-0.63) in the internal and external cohort, respectively.

CONCLUSION

The previously published results for prediction of LTP after thermal ablation of CRLM using clinical and radiomics models were not reproducible in independent internal and external validation.

CLINICAL RELEVANCE STATEMENT

Local tumour progression after thermal ablation of CRLM cannot yet be predicted with the use of CT radiomics of the ablation zone and peri-ablational rim. These results underline the importance of validation of radiomics results to test for reproducibility in independent cohorts.

KEY POINTS

• Previous research suggests CT radiomics models have the potential to predict local tumour progression after thermal ablation in colorectal liver metastases, but independent validation is lacking. • In internal and external validation, the previously published models were not able to predict local tumour progression after ablation. • Radiomics prediction models should be investigated in independent validation cohorts to check for reproducibility.

Investigation of a potential upstream harmonization based on image appearance matching to improve radiomics features robustness: a phantom study

Objective. Radiomics is a promising valuable analysis tool consisting in extracting quantitative information from medical images. However, the extracted radiomics features are too sensitive to variations in used image acquisition and reconstruction parameters. This limited robustness hinders the generalizable validity of radiomics-assisted models. Our aim is to investigate a possible harmonization strategy based on matching image quality to improve feature robustness.Approach.We acquired CT scans of a phantom with two scanners across different dose levels and percentages of Iterative Reconstruction algorithms. The detectability index was used as a comprehensive task-based image quality metric. A statistical analysis based on the Intraclass Correlation Coefficient was performed to determine if matching image quality/appearance could enhance the robustness of radiomics features extracted from the phantom images. Additionally, an Artificial Neural Network was trained on these features to automatically classify the scanner used for image acquisition.Main results.We found that the ICC of the features across protocols providing a similar detectability index improves with respect to the ICC of the features across protocols providing a different detectability index. This improvement was particularly noticeable in features relevant for distinguishing between scanners.Significance.This preliminary study demonstrates that a harmonization based on image quality/appearance matching could improve radiomics features robustness and heterogeneous protocols can be used to obtain a similar image appearance in terms of the detectability index. Thus protocols with a lower dose level could be selected to reduce the amount of radiation dose delivered to the patient and simultaneously obtain a more robust quantitative analysis.

MRI-based radiomics for predicting histology in malignant salivary gland tumors: methodology and "proof of principle"

Defining the exact histological features of salivary gland malignancies before treatment remains an unsolved problem that compromises the ability to tailor further therapeutic steps individually. Radiomics, a new methodology to extract quantitative information from medical images, could contribute to characterizing the individual cancer phenotype already before treatment in a fast and non-invasive way. Consequently, the standardization and implementation of radiomic analysis in the clinical routine work to predict histology of salivary gland cancer (SGC) could also provide improvements in clinical decision-making. In this study, we aimed to investigate the potential of radiomic features as imaging biomarker to distinguish between high grade and low-grade salivary gland malignancies. We have also investigated the effect of image and feature level harmonization on the performance of radiomic models. For this study, our dual center cohort consisted of 126 patients, with histologically proven SGC, who underwent curative-intent treatment in two tertiary oncology centers. We extracted and analyzed the radiomics features of 120 pre-therapeutic MRI images with gadolinium (T1 sequences), and correlated those with the definitive post-operative histology. In our study the best radiomic model achieved average AUC of 0.66 and balanced accuracy of 0.63. According to the results, there is significant difference between the performance of models based on MRI intensity normalized images + harmonized features and other models (p value < 0.05) which indicates that in case of dealing with heterogeneous dataset, applying the harmonization methods is beneficial. Among radiomic features minimum intensity from first order, and gray level-variance from texture category were frequently selected during multivariate analysis which indicate the potential of these features as being used as imaging biomarker. The present bicentric study presents for the first time the feasibility of implementing MR-based, handcrafted radiomics, based on T1 contrast-enhanced sequences and the ComBat harmonization method in an effort to predict the formal grading of salivary gland carcinoma with satisfactory performance.

Identification of CT radiomic features robust to acquisition and segmentation variations for improved prediction of radiotherapy-treated lung cancer patient recurrence

The primary objective of the present study was to identify a subset of radiomic features extracted from primary tumor imaged by computed tomography of early-stage non-small cell lung cancer patients, which remain unaffected by variations in segmentation quality and in computed tomography image acquisition protocol. The robustness of these features to segmentation variations was assessed by analyzing the correlation of feature values extracted from lesion volumes delineated by two annotators. The robustness to variations in acquisition protocol was evaluated by examining the correlation of features extracted from high-dose and low-dose computed tomography scans, both of which were acquired for each patient as part of the stereotactic body radiotherapy planning process. Among 106 radiomic features considered, 21 were identified as robust. An analysis including univariate and multivariate assessments was subsequently conducted to estimate the predictive performance of these robust features on the outcome of early-stage non-small cell lung cancer patients treated with stereotactic body radiation therapy. The univariate predictive analysis revealed that robust features demonstrated superior predictive potential compared to non-robust features. The multivariate analysis indicated that linear regression models built with robust features displayed greater generalization capabilities by outperforming other models in predicting the outcomes of an external validation dataset.

Deep conditional generative model for longitudinal single-slice abdominal computed tomography harmonization

Purpose

Two-dimensional single-slice abdominal computed tomography (CT) provides a detailed tissue map with high resolution allowing quantitative characterization of relationships between health conditions and aging. However, longitudinal analysis of body composition changes using these scans is difficult due to positional variation between slices acquired in different years, which leads to different organs/tissues being captured.

Approach

To address this issue, we propose C-SliceGen, which takes an arbitrary axial slice in the abdominal region as a condition and generates a pre-defined vertebral level slice by estimating structural changes in the latent space.

Results

Our experiments on 2608 volumetric CT data from two in-house datasets and 50 subjects from the 2015 Multi-Atlas Abdomen Labeling Challenge Beyond the Cranial Vault (BTCV) dataset demonstrate that our model can generate high-quality images that are realistic and similar. We further evaluate our method's capability to harmonize longitudinal positional variation on 1033 subjects from the Baltimore longitudinal study of aging dataset, which contains longitudinal single abdominal slices, and confirmed that our method can harmonize the slice positional variance in terms of visceral fat area.

Conclusion

This approach provides a promising direction for mapping slices from different vertebral levels to a target slice and reducing positional variance for single-slice longitudinal analysis. The source code is available at: https://github.com/MASILab/C-SliceGen.

Lifespan reference curves for harmonizing multi-site regional brain white matter metrics from diffusion MRI

Age-related white matter (WM) microstructure maturation and decline occur throughout the human lifespan, complementing the process of gray matter development and degeneration. Here, we create normative lifespan reference curves for global and regional WM microstructure by harmonizing diffusion MRI (dMRI)-derived data from ten public datasets (N = 40,898 subjects; age: 3-95 years; 47.6% male). We tested three harmonization methods on regional diffusion tensor imaging (DTI) based fractional anisotropy (FA), a metric of WM microstructure, extracted using the ENIGMA-DTI pipeline. ComBat-GAM harmonization provided multi-study trajectories most consistent with known WM maturation peaks. Lifespan FA reference curves were validated with test-retest data and used to assess the effect of the ApoE4 risk factor for dementia in WM across the lifespan. We found significant associations between ApoE4 and FA in WM regions associated with neurodegenerative disease even in healthy individuals across the lifespan, with regional age-by-genotype interactions. Our lifespan reference curves and tools to harmonize new dMRI data to the curves are publicly available as eHarmonize (https://github.com/ahzhu/eharmonize).

Multi-institutional prognostic modeling of survival outcomes in NSCLC patients treated with first-line immunotherapy using radiomics

BACKGROUND

Immune checkpoint inhibitors (ICIs) have emerged as one of the most promising first-line therapeutics in the management of non-small cell lung cancer (NSCLC). However, only a subset of these patients responds to ICIs, highlighting the clinical need to develop better predictive and prognostic biomarkers. This study will leverage pre-treatment imaging profiles to develop survival risk models for NSCLC patients treated with first-line immunotherapy.

METHODS

Advanced NSCLC patients (n = 149) were retrospectively identified from two institutions who were treated with first-line ICIs. Radiomics features extracted from pretreatment imaging scans were used to build the predictive models for progression-free survival (PFS) and overall survival (OS). A compendium of five feature selection methods and seven machine learning approaches were utilized to build the survival risk models. The concordance index (C-index) was used to evaluate model performance.

RESULTS

From our results, we found several combinations of machine learning algorithms and feature selection methods to achieve similar performance. K-nearest neighbourhood (KNN) with ReliefF (RL) feature selection was the best-performing model to predict PFS (C-index = 0.61 and 0.604 in discovery and validation cohorts), while XGBoost with Mutual Information (MI) feature selection was the best-performing model for OS (C-index = 0.7 and 0.655 in discovery and validation cohorts).

CONCLUSION

The results of this study highlight the importance of implementing an appropriate feature selection method coupled with a machine learning strategy to develop robust survival models. With further validation of these models on external cohorts when available, this can have the potential to improve clinical decisions by systematically analyzing routine medical images.

State of the art: radiomics and radiomics-related artificial intelligence on the road to clinical translation

Radiomics and artificial intelligence carry the promise of increased precision in oncologic imaging assessments due to the ability of harnessing thousands of occult digital imaging features embedded in conventional medical imaging data. While powerful, these technologies suffer from a number of sources of variability that currently impede clinical translation. In order to overcome this impediment, there is a need to control for these sources of variability through harmonization of imaging data acquisition across institutions, construction of standardized imaging protocols that maximize the acquisition of these features, harmonization of post-processing techniques, and big data resources to properly power studies for hypothesis testing. For this to be accomplished, it will be critical to have multidisciplinary and multi-institutional collaboration.

Optimal batch determination for improved harmonization and prognostication of multi-center PET/CT radiomics feature in head and neck cancer

Objective. To determine the optimal approach for identifying and mitigating batch effects in PET/CT radiomics features, and further improve the prognosis of patients with head and neck cancer (HNC), this study investigated the performance of three batch harmonization methods.Approach. Unsupervised harmonization identified the batch labels by K-means clustering. Supervised harmonization regarding the image acquisition factors (center, manufacturer, scanner, filter kernel) as known/given batch labels, and Combat harmonization was then implemented separately and sequentially based on the batch labels, i.e. harmonizing features among batches determined by each factor individually or harmonizing features among batches determined by multiple factors successively. Extensive experiments were conducted to predict overall survival (OS) on public PET/CT datasets that contain 800 patients from 9 centers.Main results. In the external validation cohort, results show that compared to original models without harmonization, Combat harmonization would be beneficial in OS prediction with C-index of 0.687-0.740 versus 0.684-0.767. Supervised harmonization slightly outperformed unsupervised harmonization in all models (C-index: 0.692-0.767 versus 0.684-0.750). Separate harmonization outperformed sequential harmonization in CT_m+clinic and CT_cm+clinic models with C-index of 0.752 and 0.722, respectively, while sequential harmonization involved clinical features in PET_rs+clinic model further improving the performance and achieving the highest C-index of 0.767.Significance. Optimal batch determination especially sequential harmonization for Combat holds the potential to improve the prognostic power of radiomics model in multi-center HNC dataset with PET/CT imaging.

Machine learning for differentiation of lipid-poor adrenal adenoma and subclinical pheochromocytoma based on multiphase CT imaging radiomics

BACKGROUND

There is a paucity of research investigating the application of machine learning techniques for distinguishing between lipid-poor adrenal adenoma (LPA) and subclinical pheochromocytoma (sPHEO) based on radiomic features extracted from non-contrast and dynamic contrast-enhanced computed tomography (CT) scans of the abdomen.

METHODS

We conducted a retrospective analysis of multiphase spiral CT scans, including non-contrast, arterial, venous, and delayed phases, as well as thin- and thick-thickness images from 134 patients with surgically and pathologically confirmed. A total of 52 patients with LPA and 44 patients with sPHEO were randomly assigned to training/testing sets in a 7:3 ratio. Additionally, a validation set was comprised of 22 LPA cases and 16 sPHEO cases from two other hospitals. We used 3D Slicer and PyRadiomics to segment tumors and extract radiomic features, respectively. We then applied T-test and least absolute shrinkage and selection operator (LASSO) to select features. Six binary classifiers, including K-nearest neighbor (KNN), logistic regression (LR), decision tree (DT), random forest (RF), support vector machine (SVM), and multi-layer perceptron (MLP), were employed to differentiate LPA from sPHEO. Receiver operating characteristic (ROC) curves and area under the curve (AUC) values were compared using DeLong's method.

RESULTS

All six classifiers showed good diagnostic performance for each phase and slice thickness, as well as for the entire CT data, with AUC values ranging from 0.706 to 1. Non-contrast CT densities of LPA were significantly lower than those of sPHEO (P < 0.001). However, using the optimal threshold for non-contrast CT density, sensitivity was only 0.743, specificity 0.744, and AUC 0.828. Delayed phase CT density yielded a sensitivity of 0.971, specificity of 0.641, and AUC of 0.814. In radiomics, AUC values for the testing set using non-contrast CT images were: KNN 0.919, LR 0.979, DT 0.835, RF 0.967, SVM 0.979, and MLP 0.981. In the validation set, AUC values were: KNN 0.891, LR 0.974, DT 0.891, RF 0.964, SVM 0.949, and MLP 0.979.

CONCLUSIONS

The machine learning model based on CT radiomics can accurately differentiate LPA from sPHEO, even using non-contrast CT data alone, making contrast-enhanced CT unnecessary for diagnosing LPA and sPHEO.

A radiomics-based logistic regression model for the assessment of emphysema severity

Introduction

The aim of this study is to develop a model that differentiates between the radiological patterns of severe and mild emphysema using radiomics parameters, as well as to examine the parameters included in the model.

Materials and Methods

Over the last 12 months, a total of 354 patients were screened based on the presence of terms such as “Fleischner”, “CLE”, and “centriacinar” in their thoracic CT reports, culminating in a study population of 82 patients. The study population was divided into Group 1 (Fleischner mild and moderate; n= 45) and Group 2 (Fleischner confluent and advanced destructive; n= 37). Volumetric segmentation was performed, focusing on the upper lobe segments of both lungs. From these segmented volumes, radiomics parameters including shape, size, first-order, and second-order features were calculated. The best model parameters were selected based on the Bayesian Information Criterion and further optimized through grid search. The final model was tested using 1000 iterations of bootstrap resampling.

Results

In the training set, performance metrics were calculated with a sensitivity of 0.862, specificity of 0.870, accuracy of 0.863, and AUC of 0.910. Correspondingly, in the test set, these values were sensitivity= 0.848; specificity= 0.865; accuracy= 0.857; and AUC= 0.907.

Conclusion

The logistic regression model, composed of radiomics parameters and trained on a limited number of cases, effectively differentiated between mild and severe radiological patterns of emphysema using computed tomography images.

Combining Molecular and Radiomic Features for Risk Assessment in Breast Cancer

Breast cancer risk is highly variable within the population and current research is leading the shift toward personalized medicine. By accurately assessing an individual woman's risk, we can reduce the risk of over/undertreatment by preventing unnecessary procedures or by elevating screening procedures. Breast density measured from conventional mammography has been established as one of the most dominant risk factors for breast cancer; however, it is currently limited by its ability to characterize more complex breast parenchymal patterns that have been shown to provide additional information to strengthen cancer risk models. Molecular factors ranging from high penetrance, or high likelihood that a mutation will show signs and symptoms of the disease, to combinations of gene mutations with low penetrance have shown promise for augmenting risk assessment. Although imaging biomarkers and molecular biomarkers have both individually demonstrated improved performance in risk assessment, few studies have evaluated them together. This review aims to highlight the current state of the art in breast cancer risk assessment using imaging and genetic biomarkers.

ComBat Harmonization of Myocardial Radiomic Features Sensitive to Cardiac MRI Acquisition Parameters

Purpose

To investigate the effect of ComBat harmonization methods on the robustness of cardiac MRI-derived radiomic features to variations in imaging parameters.

Materials and Methods

This Health Insurance Portability and Accountability Act-compliant retrospective study used a publicly available data set of 11 healthy controls (mean age, 33 years ± 16 [SD]; six men) and five patients (mean age, 52 years ± 16; four men). A single midventricular short-axis section was acquired with 3-T MRI using cine balanced steady-state free precision, T1-weighted, T2-weighted, T1 mapping, and T2 mapping imaging sequences. Each sequence was acquired using baseline parameters and after variations in flip angle, spatial resolution, section thickness, and parallel imaging. Image registration was performed for all sequences at a per-individual level. Manual myocardial contouring was performed, and 1652 radiomic features per sequence were extracted using baseline and variations in imaging parameters. Radiomic feature stability to change in imaging parameters was assessed using Cohen d sensitivity. The stability of radiomic features was assessed both without and after ComBat harmonization of radiomic features. Three ComBat methods were studied: parametric, nonparametric, and Gaussian mixture model (GMM).

Results

For all sequences combined, 51.4% of features were robust to changes in imaging parameters when no ComBat method was applied. ComBat harmonization substantially increased the number of stable features to 95.1% (95% CI: 94.9, 95.3) when parametric ComBat was used and 90.9% (95% CI: 90.6, 91.2) when nonparametric ComBat was used. GMM combat resulted in only 52.6% stable features.

Conclusion

ComBat harmonization improved the stability of radiomic features to changes in imaging parameters across all cardiac MRI sequences.Keywords: Cardiac MRI, Radiomics, ComBat, Harmonization Supplemental material is available for this article. © RSNA, 2023.

Generalization optimizing machine learning to improve CT scan radiomics and assess immune checkpoint inhibitors' response in non-small cell lung cancer: a multicenter cohort study

Background

Recent developments in artificial intelligence suggest that radiomics may represent a promising non-invasive biomarker to predict response to immune checkpoint inhibitors (ICIs). Nevertheless, validation of radiomics algorithms in independent cohorts remains a challenge due to variations in image acquisition and reconstruction. Using radiomics, we investigated the importance of scan normalization as part of a broader machine learning framework to enable model external generalizability to predict ICI response in non-small cell lung cancer (NSCLC) patients across different centers.

Methods

Radiomics features were extracted and compared from 642 advanced NSCLC patients on pre-ICI scans using established open-source PyRadiomics and a proprietary DeepRadiomics deep learning technology. The population was separated into two groups: a discovery cohort of 512 NSCLC patients from three academic centers and a validation cohort that included 130 NSCLC patients from a fourth center. We harmonized images to account for variations in reconstruction kernel, slice thicknesses, and device manufacturers. Multivariable models, evaluated using cross-validation, were used to estimate the predictive value of clinical variables, PD-L1 expression, and PyRadiomics or DeepRadiomics for progression-free survival at 6 months (PFS-6).

Results

The best prognostic factor for PFS-6, excluding radiomics features, was obtained with the combination of Clinical + PD-L1 expression (AUC = 0.66 in the discovery and 0.62 in the validation cohort). Without image harmonization, combining Clinical + PyRadiomics or DeepRadiomics delivered an AUC = 0.69 and 0.69, respectively, in the discovery cohort, but dropped to 0.57 and 0.52, in the validation cohort. This lack of generalizability was consistent with observations in principal component analysis clustered by CT scan parameters. Subsequently, image harmonization eliminated these clusters. The combination of Clinical + DeepRadiomics reached an AUC = 0.67 and 0.63 in the discovery and validation cohort, respectively. Conversely, the combination of Clinical + PyRadiomics failed generalizability validations, with AUC = 0.66 and 0.59.

Conclusion

We demonstrated that a risk prediction model combining Clinical + DeepRadiomics was generalizable following CT scan harmonization and machine learning generalization methods. These results had similar performances to routine oncology practice using Clinical + PD-L1. This study supports the strong potential of radiomics as a future non-invasive strategy to predict ICI response in advanced NSCLC.

Systematic Review, Meta-Analysis and Radiomics Quality Score Assessment of CT Radiomics-Based Models Predicting Tumor EGFR Mutation Status in Patients with Non-Small-Cell Lung Cancer

Assessment of the quality and current performance of computed tomography (CT) radiomics-based models in predicting epidermal growth factor receptor (EGFR) mutation status in patients with non-small-cell lung carcinoma (NSCLC). Two medical literature databases were systematically searched, and articles presenting original studies on CT radiomics-based models for predicting EGFR mutation status were retrieved. Forest plots and related statistical tests were performed to summarize the model performance and inter-study heterogeneity. The methodological quality of the selected studies was assessed via the Radiomics Quality Score (RQS). The performance of the models was evaluated using the area under the curve (ROC AUC). The range of the Risk RQS across the selected articles varied from 11 to 24, indicating a notable heterogeneity in the quality and methodology of the included studies. The average score was 15.25, which accounted for 42.34% of the maximum possible score. The pooled Area Under the Curve (AUC) value was 0.801, indicating the accuracy of CT radiomics-based models in predicting the EGFR mutation status. CT radiomics-based models show promising results as non-invasive alternatives for predicting EGFR mutation status in NSCLC patients. However, the quality of the studies using CT radiomics-based models varies widely, and further harmonization and prospective validation are needed before the generalization of these models.

Image harmonization: A review of statistical and deep learning methods for removing batch effects and evaluation metrics for effective harmonization

Magnetic resonance imaging and computed tomography from multiple batches (e.g. sites, scanners, datasets, etc.) are increasingly used alongside complex downstream analyses to obtain new insights into the human brain. However, significant confounding due to batch-related technical variation, called batch effects, is present in this data; direct application of downstream analyses to the data may lead to biased results. Image harmonization methods seek to remove these batch effects and enable increased generalizability and reproducibility of downstream results. In this review, we describe and categorize current approaches in statistical and deep learning harmonization methods. We also describe current evaluation metrics used to assess harmonization methods and provide a standardized framework to evaluate newly-proposed methods for effective harmonization and preservation of biological information. Finally, we provide recommendations to end-users to advocate for more effective use of current methods and to methodologists to direct future efforts and accelerate development of the field.

Enhancing fetal alcohol spectrum disorders diagnosis with a classifier based on the intracerebellar gradient of volumetric undersizing

In fetal alcohol spectrum disorders (FASD), brain growth deficiency is a hallmark of subjects both with fetal alcohol syndrome (FAS) and with non-syndromic FASD (NS-FASD, i.e., those without specific diagnostic features). However, although the cerebellum was suggested to be more severely undersized than the rest of the brain, it has not yet been given a specific place in the FASD diagnostic criteria where neuroanatomical features still count for little if anything in diagnostic specificity. We applied a combination of cerebellar segmentation tools on a 1.5 T 3DT1 brain MRI dataset from a monocentric population of 89 FASD (52 FAS, 37 NS-FASD) and 126 typically developing controls (6-20 years old), providing 8 volumes: cerebellum, vermis and 3 lobes (anterior, posterior, inferior), plus total brain volume. After adjustment of confounders, the allometric scaling relationship between these cerebellar volumes (V_i ) and the total brain or cerebellum volume (V_t ) was fitted (V_i  = bV_t ^a ), and the effect of group (FAS, control) on allometric scaling was evaluated. We then estimated for each cerebellar volume in the FAS population the deviation from the typical scaling (_v DTS) learned in the controls. Lastly, we trained and tested two classifiers to discriminate FAS from controls, one based on the total cerebellum _v DTS only, the other based on all the cerebellar _v DTS, comparing their performance both in the FAS and the NS-FASD group. Allometric scaling was significantly different between FAS and control group for all the cerebellar volumes (p < .001). We confirmed the excess of total cerebellum volume deficit (_v DTS = -10.6%) and revealed an antero-inferior-posterior gradient of volumetric undersizing in the hemispheres (-12.4%, 1.1%, 2.0%, repectively) and the vermis (-16.7%, -9.2%, -8.6%, repectively). The classifier based on the intracerebellar gradient of _v DTS performed more efficiently than the one based on total cerebellum _v DTS only (AUC = 92% vs. 82%, p = .001). Setting a high probability threshold for >95% specificity of the classifiers, the gradient-based classifier identified 35% of the NS-FASD to have a FAS cerebellar phenotype, compared to 11% with the cerebellum-only classifier (p_FISHER  = 0.027). In a large series of FASD, this study details the volumetric undersizing within the cerebellum at the lobar and vermian level using allometric scaling, revealing an anterior-inferior-posterior gradient of vulnerability to prenatal alcohol exposure. It also strongly suggests that this intracerebellar gradient of volumetric undersizing may be a reliable neuroanatomical signature of FAS that could be used to improve the specificity of the diagnosis of NS-FASD.

Better Together: Data Harmonization and Cross-Study Analysis of Abdominal MRI Data From UK Biobank and the German National Cohort

OBJECTIVES

The UK Biobank (UKBB) and German National Cohort (NAKO) are among the largest cohort studies, capturing a wide range of health-related data from the general population, including comprehensive magnetic resonance imaging (MRI) examinations. The purpose of this study was to demonstrate how MRI data from these large-scale studies can be jointly analyzed and to derive comprehensive quantitative image-based phenotypes across the general adult population.

MATERIALS AND METHODS

Image-derived features of abdominal organs (volumes of liver, spleen, kidneys, and pancreas; volumes of kidney hilum adipose tissue; and fat fractions of liver and pancreas) were extracted from T1-weighted Dixon MRI data of 17,996 participants of UKBB and NAKO based on quality-controlled deep learning generated organ segmentations. To enable valid cross-study analysis, we first analyzed the data generating process using methods of causal discovery. We subsequently harmonized data from UKBB and NAKO using the ComBat approach for batch effect correction. We finally performed quantile regression on harmonized data across studies providing quantitative models for the variation of image-derived features stratified for sex and dependent on age, height, and weight.

RESULTS

Data from 8791 UKBB participants (49.9% female; age, 63 ± 7.5 years) and 9205 NAKO participants (49.1% female, age: 51.8 ± 11.4 years) were analyzed. Analysis of the data generating process revealed direct effects of age, sex, height, weight, and the data source (UKBB vs NAKO) on image-derived features. Correction of data source-related effects resulted in markedly improved alignment of image-derived features between UKBB and NAKO. Cross-study analysis on harmonized data revealed comprehensive quantitative models for the phenotypic variation of abdominal organs across the general adult population.

CONCLUSIONS

Cross-study analysis of MRI data from UKBB and NAKO as proposed in this work can be helpful for future joint data analyses across cohorts linking genetic, environmental, and behavioral risk factors to MRI-derived phenotypes and provide reference values for clinical diagnostics.

Artificial Intelligence in CT and MR Imaging for Oncological Applications

Cancer care increasingly relies on imaging for patient management. The two most common cross-sectional imaging modalities in oncology are computed tomography (CT) and magnetic resonance imaging (MRI), which provide high-resolution anatomic and physiological imaging. Herewith is a summary of recent applications of rapidly advancing artificial intelligence (AI) in CT and MRI oncological imaging that addresses the benefits and challenges of the resultant opportunities with examples. Major challenges remain, such as how best to integrate AI developments into clinical radiology practice, the vigorous assessment of quantitative CT and MR imaging data accuracy, and reliability for clinical utility and research integrity in oncology. Such challenges necessitate an evaluation of the robustness of imaging biomarkers to be included in AI developments, a culture of data sharing, and the cooperation of knowledgeable academics with vendor scientists and companies operating in radiology and oncology fields. Herein, we will illustrate a few challenges and solutions of these efforts using novel methods for synthesizing different contrast modality images, auto-segmentation, and image reconstruction with examples from lung CT as well as abdome, pelvis, and head and neck MRI. The imaging community must embrace the need for quantitative CT and MRI metrics beyond lesion size measurement. AI methods for the extraction and longitudinal tracking of imaging metrics from registered lesions and understanding the tumor environment will be invaluable for interpreting disease status and treatment efficacy. This is an exciting time to work together to move the imaging field forward with narrow AI-specific tasks. New AI developments using CT and MRI datasets will be used to improve the personalized management of cancer patients.

DeepComBat: A Statistically Motivated, Hyperparameter-Robust, Deep Learning Approach to Harmonization of Neuroimaging Data

Neuroimaging data from multiple batches (i.e. acquisition sites, scanner manufacturer, datasets, etc.) are increasingly necessary to gain new insights into the human brain. However, multi-batch data, as well as extracted radiomic features, exhibit pronounced technical artifacts across batches. These batch effects introduce confounding into the data and can obscure biological effects of interest, decreasing the generalizability and reproducibility of findings. This is especially true when multi-batch data is used alongside complex downstream analysis models, such as machine learning methods. Image harmonization methods seeking to remove these batch effects are important for mitigating these issues; however, significant multivariate batch effects remain in the data following harmonization by current state-of-the-art statistical and deep learning methods. We present DeepCombat, a deep learning harmonization method based on a conditional variational autoencoder architecture and the ComBat harmonization model. DeepCombat learns and removes subject-level batch effects by accounting for the multivariate relationships between features. Additionally, DeepComBat relaxes a number of strong assumptions commonly made by previous deep learning harmonization methods and is empirically robust across a wide range of hyperparameter choices. We apply this method to neuroimaging data from a large cognitive-aging cohort and find that DeepCombat outperforms existing methods, as assessed by a battery of machine learning methods, in removing scanner effects from cortical thickness measurements while preserving biological heterogeneity. Additionally, DeepComBat provides a new perspective for statistically-motivated deep learning harmonization methods.

A radiomics model based on aortic computed tomography angiography: the impact on predicting the prognosis of patients with aortic intramural hematoma (IMH)

Background

The prognosis of aortic intramural hematoma (IMH) is unpredictable, but computed tomography angiography (CTA) plays an important role of high diagnostic performance in the initial diagnosis and during follow-up of patients. In this study, we investigated the value of a radiomics model based on aortic CTA for predicting the prognosis of patients with medically treated IMH.

Method

A total of 120 patients with IMH were enrolled in this study. The follow-up duration ranged from 32 to 1,346 days (median 232 days). Progression of these patients was classified as follows: destabilization, which refers to deterioration in the aortic condition, including significant increases in the thickness of the IMH, the progression of IMH to a penetrating aortic ulcer (PAU), aortic dissection (AD), or rupture; or stabilization, which refers to an unchanged appearance or a decrease in the size or disappearance of the IMH. The patients were divided into a training cohort (n=84) and a validation cohort (n=36). Six different machine learning classifiers were applied: random forest (RF), K-nearest neighbor (KNN), Gaussian Naive Bayes, decision tree, logistic regression, and support vector machine (SVM). The clinical-radiomics combined nomogram model was established by multivariate logistic regression. The area under the receiver operating characteristic (ROC) curve (AUC) was implemented to evaluate the discrimination performance of the models. The calibration curves and Hosmer-Lemeshow test were used for evaluating model calibration. DeLong's test was performed to compare the AUC performance of models.

Results

Among all of the patients, 60 patients showed destabilization and 60 patients remained stable. A total of 12 radiomic features were retained after application of the least absolute shrinkage and selection operator (LASSO). These features were used for the machine learning model construction. The SVM-radial basis function (SVM-RBF) model obtained the best performance with an AUC of 0.765 (95% CI, 0.593-0.906). In the validation cohort, the combined clinical-radiomics model [AUC =0.787; 95% confidence interval (CI), 0.619-0.923] showed a significantly higher performance than did the clinical model (AUC =0.596; 95% CI, 0.413-0.796; P=0.021) and had a similar performance to the radiomics model (AUC =0.765; 95% CI, 0.589-0.906; P=0.672).

Conclusions

A quantitative nomogram based on radiomic features of CTA images can be used to predict disease progression in patients with IMH and may help improve clinical decision-making.

Improved generalized ComBat methods for harmonization of radiomic features

Radiomic approaches in precision medicine are promising, but variation associated with image acquisition factors can result in severe biases and low generalizability. Multicenter datasets used in these studies are often heterogeneous in multiple imaging parameters and/or have missing information, resulting in multimodal radiomic feature distributions. ComBat is a promising harmonization tool, but it only harmonizes by single/known variables and assumes standardized input data are normally distributed. We propose a procedure that sequentially harmonizes for multiple batch effects in an optimized order, called OPNested ComBat. Furthermore, we propose to address bimodality by employing a Gaussian Mixture Model (GMM) grouping considered as either a batch variable (OPNested + GMM) or as a protected clinical covariate (OPNested - GMM). Methods were evaluated on features extracted with CapTK and PyRadiomics from two public lung computed tomography (CT) datasets. We found that OPNested ComBat improved harmonization performance over standard ComBat. OPNested + GMM ComBat exhibited the best harmonization performance but the lowest predictive performance, while OPNested - GMM ComBat showed poorer harmonization performance, but the highest predictive performance. Our findings emphasize that improved harmonization performance is no guarantee of improved predictive performance, and that these methods show promise for superior standardization of datasets heterogeneous in multiple or unknown imaging parameters and greater generalizability.

Harmonization Strategies in Multicenter MRI-Based Radiomics

Radiomics analysis is a powerful tool aiming to provide diagnostic and prognostic patient information directly from images that are decoded into handcrafted features, comprising descriptors of shape, size and textural patterns. Although radiomics is gaining momentum since it holds great promise for accelerating digital diagnostics, it is susceptible to bias and variation due to numerous inter-patient factors (e.g., patient age and gender) as well as inter-scanner ones (different protocol acquisition depending on the scanner center). A variety of image and feature based harmonization methods has been developed to compensate for these effects; however, to the best of our knowledge, none of these techniques has been established as the most effective in the analysis pipeline so far. To this end, this review provides an overview of the challenges in optimizing radiomics analysis, and a concise summary of the most relevant harmonization techniques, aiming to provide a thorough guide to the radiomics harmonization process.

Long-term effect of childhood trauma: Role of inflammation and white matter in mood disorders

Bipolar disorder (BD) and major depressive disorder (MDD) are severe psychiatric illnesses that share among their environmental risk factors the exposure to adverse childhood experiences (ACE). Exposure to ACE has been associated with long-term changes in brain structure and the immune response. In the lasts decades, brain abnormalities including alterations of white matter (WM) microstructure and higher levels of peripheral immune/inflammatory markers have been reported in BD and MDD and an association between inflammation and WM microstructure has been shown. However, differences in these measures have been reported by comparing the two diagnostic groups. The aim of the present study was to investigate the interplay between ACE, inflammation, and WM in BD and MDD. We hypothesize that inflammation will mediate the association between ACE and WM and that this will be different in the two groups. A sample of 200 patients (100 BD, 100 MDD) underwent 3T MRI scan and ACE assessment through Childhood Trauma Questionnaire. A subgroup of 130 patients (75 MDD and 55 BD) underwent blood sampling for the assessment of immune/inflammatory markers. We observed that ACE associated with higher peripheral levels of IL-2, IL-17, bFGF, IFN-γ, TNF-α, CCL3, CCL4, CCL5, and PDGF-BB only in the BD group. Further, higher levels of CCL3 and IL-2 associated with lower FA in BD. ACE were found to differently affect WM microstructure in the two diagnostic groups and to be negatively associated with FA and AD in BD patients. Mediation analyses showed a significant indirect effect of ACE on WM microstructure mediated by IL-2. Our findings suggest that inflammation may mediate the detrimental effect of early experiences on brain structure and different mechanisms underlying brain alterations in BD and MDD.

matRadiomics: A Novel and Complete Radiomics Framework, from Image Visualization to Predictive Model

Radiomics aims to support clinical decisions through its workflow, which is divided into: (i) target identification and segmentation, (ii) feature extraction, (iii) feature selection, and (iv) model fitting. Many radiomics tools were developed to fulfill the steps mentioned above. However, to date, users must switch different software to complete the radiomics workflow. To address this issue, we developed a new free and user-friendly radiomics framework, namely matRadiomics, which allows the user: (i) to import and inspect biomedical images, (ii) to identify and segment the target, (iii) to extract the features, (iv) to reduce and select them, and (v) to build a predictive model using machine learning algorithms. As a result, biomedical images can be visualized and segmented and, through the integration of Pyradiomics into matRadiomics, radiomic features can be extracted. These features can be selected using a hybrid descriptive–inferential method, and, consequently, used to train three different classifiers: linear discriminant analysis, k-nearest neighbors, and support vector machines. Model validation is performed using k-fold cross-Validation and k-fold stratified cross-validation. Finally, the performance metrics of each model are shown in the graphical interface of matRadiomics. In this study, we discuss the workflow, architecture, application, future development of matRadiomics, and demonstrate its working principles in a real case study with the aim of establishing a reference standard for the whole radiomics analysis, starting from the image visualization up to the predictive model implementation.