From explainable to interpretable deep learning for natural language processing in healthcare: How far from reality?

Deep learning (DL) has substantially enhanced natural language processing (NLP) in healthcare research. However, the increasing complexity of DL-based NLP necessitates transparent model interpretability, or at least explainability, for reliable decision-making. This work presents a thorough scoping review of explainable and interpretable DL in healthcare NLP. The term "eXplainable and Interpretable Artificial Intelligence" (XIAI) is introduced to distinguish XAI from IAI. Different models are further categorized based on their functionality (model-, input-, output-based) and scope (local, global). Our analysis shows that attention mechanisms are the most prevalent emerging IAI technique. The use of IAI is growing, distinguishing it from XAI. The major challenges identified are that most XIAI does not explore "global" modelling processes, the lack of best practices, and the lack of systematic evaluation and benchmarks. One important opportunity is to use attention mechanisms to enhance multi-modal XIAI for personalized medicine. Additionally, combining DL with causal logic holds promise. Our discussion encourages the integration of XIAI in Large Language Models (LLMs) and domain-specific smaller models. In conclusion, XIAI adoption in healthcare requires dedicated in-house expertise. Collaboration with domain experts, end-users, and policymakers can lead to ready-to-use XIAI methods across NLP and medical tasks. While challenges exist, XIAI techniques offer a valuable foundation for interpretable NLP algorithms in healthcare.

A novel deep learning method for large-scale analysis of bone marrow adiposity using UK Biobank Dixon MRI data.

Background: Bone marrow adipose tissue (BMAT) represents > 10% fat mass in healthy humans and can be measured by magnetic resonance imaging (MRI) as the bone marrow fat fraction (BMFF). Human MRI studies have identified several diseases associated with BMFF but have been relatively small scale. Population-scale studies therefore have huge potential to reveal BMAT's true clinical relevance. The UK Biobank (UKBB) is undertaking MRI of 100,000 participants, providing the ideal opportunity for such advances. Objective: To establish deep learning for high-throughput multi-site BMFF analysis from UKBB MRI data. Materials and methods: We studied males and females aged 60-69. Bone marrow (BM) segmentation was automated using a new lightweight attention-based 3D U-Net convolutional neural network that improved segmentation of small structures from large volumetric data. Using manual segmentations from 61-64 subjects, the models were trained to segment four BM regions of interest: the spine (thoracic and lumbar vertebrae), femoral head, total hip and femoral diaphysis. Models were tested using a further 10-12 datasets per region and validated using datasets from 729 UKBB participants. BMFF was then quantified and pathophysiological characteristics assessed, including site- and sex-dependent differences and the relationships with age, BMI, bone mineral density, peripheral adiposity, and osteoporosis. Results: Model accuracy matched or exceeded that for conventional U-Nets, yielding Dice scores of 91.2% (spine), 94.5% (femoral head), 91.2% (total hip) and 86.6% (femoral diaphysis). One case of severe scoliosis prevented segmentation of the spine, while one case of Non-Hodgkin Lymphoma prevented segmentation of the spine, femoral head and total hip because of T2 signal depletion; however, successful segmentation was not disrupted by any other pathophysiological variables. The resulting BMFF measurements confirmed expected relationships between BMFF and age, sex and bone density, and identified new site- and sex-specific characteristics. Conclusions: We have established a new deep learning method for accurate segmentation of small structures from large volumetric data, allowing high-throughput multi-site BMFF measurement in the UKBB. Our findings reveal new pathophysiological insights, highlighting the potential of BMFF as a novel clinical biomarker. Applying our method across the full UKBB cohort will help to reveal the impact of BMAT on human health and disease.

Is Attention all You Need in Medical Image Analysis? A Review.

Medical imaging is a key component in clinical diagnosis, treatment planning and clinical trial design, accounting for almost 90% of all healthcare data. CNNs achieved performance gains in medical image analysis (MIA) over the last years. CNNs can efficiently model local pixel interactions and be trained on small-scale MI data. Despite their important advances, typical CNN have relatively limited capabilities in modelling "global" pixel interactions, which restricts their generalisation ability to understand out-of-distribution data with different "global" information. The recent progress of Artificial Intelligence gave rise to Transformers, which can learn global relationships from data. However, full Transformer models need to be trained on large-scale data and involve tremendous computational complexity. Attention and Transformer compartments ("Transf/Attention") which can well maintain properties for modelling global relationships, have been proposed as lighter alternatives of full Transformers. Recently, there is an increasing trend to co-pollinate complementary local-global properties from CNN and Transf/Attention architectures, which led to a new era of hybrid models. The past years have witnessed substantial growth in hybrid CNN-Transf/Attention models across diverse MIA problems. In this systematic review, we survey existing hybrid CNN-Transf/Attention models, review and unravel key architectural designs, analyse breakthroughs, and evaluate current and future opportunities as well as challenges. We also introduced an analysis framework on generalisation opportunities of scientific and clinical impact, based on which new data-driven domain generalisation and adaptation methods can be stimulated.

Large-scale deep learning analysis to identify adult patients at risk for combined and common variable immunodeficiencies.

BACKGROUND: Primary immunodeficiency (PI) is a group of heterogeneous disorders resulting from immune system defects. Over 70% of PI is undiagnosed, leading to increased mortality, co-morbidity and healthcare costs. Among PI disorders, combined immunodeficiencies (CID) are characterized by complex immune defects. Common variable immunodeficiency (CVID) is among the most common types of PI. In light of available treatments, it is critical to identify adult patients at risk for CID and CVID, before the development of serious morbidity and mortality. METHODS: We developed a deep learning-based method (named "TabMLPNet") to analyze clinical history from nationally representative medical claims from electronic health records (Optum® data, covering all US), evaluated in the setting of identifying CID/CVID in adults. Further, we revealed the most important CID/CVID-associated antecedent phenotype combinations. Four large cohorts were generated: a total of 47,660 PI cases and (1:1 matched) controls. RESULTS: The sensitivity/specificity of TabMLPNet modeling ranges from 0.82-0.88/0.82-0.85 across cohorts. Distinctive combinations of antecedent phenotypes associated with CID/CVID are identified, consisting of respiratory infections/conditions, genetic anomalies, cardiac defects, autoimmune diseases, blood disorders and malignancies, which can possibly be useful to systematize the identification of CID and CVID. CONCLUSIONS: We demonstrated an accurate method in terms of CID and CVID detection evaluated on large-scale medical claims data. Our predictive scheme can potentially lead to the development of new clinical insights and expanded guidelines for identification of adult patients at risk for CID and CVID as well as be used to improve patient outcomes on population level.

Primary immunodeficiencies (PI) are disorders that weaken the immune system, increasing the incident of life-threatening infections, organ damage and the development of cancer and autoimmune diseases. Although PI is estimated to affect 1-2% of the global population, 70-90% of these patients remain undiagnosed. Many patients are diagnosed during adulthood, after other serious diseases have already developed. We developed a computational method to analyze the clinical history from a large group of people with and without PI. We focused on combined (CID) and common variable immunodeficiency (CVID), which are among the least studied and most common PI subtypes, respectively. We could identify people with CID or CVID and combinations of diseases and symptoms which could make it easier to identify CID or CVID. Our method could be used to more readily identify adults at risk of CID or CVID, enabling treatment to start earlier and their long-term health to be improved.

Less Is More: Unsupervised Mask-Guided Annotated CT Image Synthesis With Minimum Manual Segmentations.

As a pragmatic data augmentation tool, data synthesis has generally returned dividends in performance for deep learning based medical image analysis. However, generating corresponding segmentation masks for synthetic medical images is laborious and subjective. To obtain paired synthetic medical images and segmentations, conditional generative models that use segmentation masks as synthesis conditions were proposed. However, these segmentation mask-conditioned generative models still relied on large, varied, and labeled training datasets, and they could only provide limited constraints on human anatomical structures, leading to unrealistic image features. Moreover, the invariant pixel-level conditions could reduce the variety of synthetic lesions and thus reduce the efficacy of data augmentation. To address these issues, in this work, we propose a novel strategy for medical image synthesis, namely Unsupervised Mask (UM)-guided synthesis, to obtain both synthetic images and segmentations using limited manual segmentation labels. We first develop a superpixel based algorithm to generate unsupervised structural guidance and then design a conditional generative model to synthesize images and annotations simultaneously from those unsupervised masks in a semi-supervised multi-task setting. In addition, we devise a multi-scale multi-task Fréchet Inception Distance (MM-FID) and multi-scale multi-task standard deviation (MM-STD) to harness both fidelity and variety evaluations of synthetic CT images. With multiple analyses on different scales, we could produce stable image quality measurements with high reproducibility. Compared with the segmentation mask guided synthesis, our UM-guided synthesis provided high-quality synthetic images with significantly higher fidelity, variety, and utility ( by Wilcoxon Signed Ranked test).

Bimetallic gold-platinum nanoparticles as a drug delivery system coated with a new drug to target glioblastoma.

A drug delivery nanosystem of noble bimetallic nanoparticles (NPs) which consists of Au NPs capped with Pt NPs (Au@Pt NPs) is constructed and functionalised with a quinazoline based small molecule (Au@Pt@Q NPs), acting as a theranostic agent against glioblastoma. Two different hydrothermal synthetic procedures for bimetallic Au@Pt NPs are presented and the resulting nanostructures are fully characterised by means of spectroscopic and microscopic methods. The imaging and targeting capacity of the new drug delivery system is assessed through fluorescent optical microscopy and cytotoxicity evaluations. The constructed Au@Pt NPs consist a monodispersed colloidal solution of 25 nm with photoluminescent, fluorescent and X-Ray absorption properties that confirm their diagnostic potential. Haemolysis testing demonstrated that Au@Pt NPs are biocompatible and fluorescent microscopy confirmed their entering the cells. Cytological evaluation of the NPs through MTT assay showed that they do not inhibit the proliferation of control cell line HEK293, whereas they are toxic in U87MG, U251 and D54 glioblastoma cell lines; rendering them selective targeting agents for treating glioblastoma.

Unsupervised Image Registration towards Enhancing Performance and Explainability in Cardiac and Brain Image Analysis.

Magnetic Resonance Imaging (MRI) typically recruits multiple sequences (defined here as "modalities"). As each modality is designed to offer different anatomical and functional clinical information, there are evident disparities in the imaging content across modalities. Inter- and intra-modality affine and non-rigid image registration is an essential medical image analysis process in clinical imaging, as for example before imaging biomarkers need to be derived and clinically evaluated across different MRI modalities, time phases and slices. Although commonly needed in real clinical scenarios, affine and non-rigid image registration is not extensively investigated using a single unsupervised model architecture. In our work, we present an unsupervised deep learning registration methodology that can accurately model affine and non-rigid transformations, simultaneously. Moreover, inverse-consistency is a fundamental inter-modality registration property that is not considered in deep learning registration algorithms. To address inverse consistency, our methodology performs bi-directional cross-modality image synthesis to learn modality-invariant latent representations, and involves two factorised transformation networks (one per each encoder-decoder channel) and an inverse-consistency loss to learn topology-preserving anatomical transformations. Overall, our model (named "FIRE") shows improved performances against the reference standard baseline method (i.e., Symmetric Normalization implemented using the ANTs toolbox) on multi-modality brain 2D and 3D MRI and intra-modality cardiac 4D MRI data experiments. We focus on explaining model-data components to enhance model explainability in medical image registration. On computational time experiments, we show that the FIRE model performs on a memory-saving mode, as it can inherently learn topology-preserving image registration directly in the training phase. We therefore demonstrate an efficient and versatile registration technique that can have merit in multi-modal image registrations in the clinical setting.

Assessment of stunned and viable myocardium using manganese-enhanced MRI.

OBJECTIVE: In a proof-of-concept study, to quantify myocardial viability in patients with acute myocardial infarction using manganese-enhanced MRI (MEMRI), a measure of intracellular calcium handling. METHODS: Healthy volunteers (n=20) and patients with ST-elevation myocardial infarction (n=20) underwent late gadolinium enhancement (LGE) using gadobutrol and MEMRI using manganese dipyridoxyl diphosphate. Patients were scanned ≤7 days after reperfusion and rescanned after 3 months. Differential manganese uptake was described using a two-compartment model. RESULTS: After manganese administration, healthy control and remote non-infarcted myocardium showed a sustained 25% reduction in T1 values (mean reductions, 288±34 and 281±12 ms). Infarcted myocardium demonstrated less T1 shortening than healthy control or remote myocardium (1157±74 vs 859±36 and 835±28 ms; both p<0.0001) with intermediate T1 values (1007±31 ms) in peri-infarct regions. Compared with LGE, MEMRI was more sensitive in detecting dysfunctional myocardium (dysfunctional fraction 40.5±11.9 vs 34.9%±13.9%; p=0.02) and tracked more closely with abnormal wall motion (r2=0.72 vs 0.55; p<0.0001). Kinetic modelling showed reduced myocardial manganese influx between remote, peri-infarct and infarct regions, enabling absolute discrimination of infarcted myocardium. After 3 months, manganese uptake increased in peri-infarct regions (16.5±3.5 vs 22.8±3.5 mL/100 g/min, p<0.0001), but not the remote (23.3±2.8 vs 23.0±3.2 mL/100 g/min, p=0.8) or infarcted (11.5±3.7 vs 14.0±1.2 mL/100 g/min, p>0.1) myocardium. CONCLUSIONS: Through visualisation of intracellular calcium handling, MEMRI accurately differentiates infarcted, stunned and viable myocardium, and correlates with myocardial dysfunction better than LGE. MEMRI holds major promise in directly assessing myocardial viability, function and calcium handling across a range of cardiac diseases. TRIAL REGISTRATION NUMBERS: NCT03607669; EudraCT number 2016-003782-25.

DiCyc: GAN-based deformation invariant cross-domain information fusion for medical image synthesis.

Cycle-consistent generative adversarial network (CycleGAN) has been widely used for cross-domain medical image synthesis tasks particularly due to its ability to deal with unpaired data. However, most CycleGAN-based synthesis methods cannot achieve good alignment between the synthesized images and data from the source domain, even with additional image alignment losses. This is because the CycleGAN generator network can encode the relative deformations and noises associated to different domains. This can be detrimental for the downstream applications that rely on the synthesized images, such as generating pseudo-CT for PET-MR attenuation correction. In this paper, we present a deformation invariant cycle-consistency model that can filter out these domain-specific deformation. The deformation is globally parameterized by thin-plate-spline (TPS), and locally learned by modified deformable convolutional layers. Robustness to domain-specific deformations has been evaluated through experiments on multi-sequence brain MR data and multi-modality abdominal CT and MR data. Experiment results demonstrated that our method can achieve better alignment between the source and target data while maintaining superior image quality of signal compared to several state-of-the-art CycleGAN-based methods.

Synthesis and characterization of modified magnetic nanoparticles as theranostic agents: in vitro safety assessment in healthy cells.

Over the past few decades nanotechnology has paved its way into cancer treatment procedures with the use of nanoparticles (NPs) for contrast media and therapeutic agents. Iron based NPs are the most investigated since they can be used for drug delivery, imaging and when magnetically activate employed as local heat sources in cancer hyperthermia. In this work, was performed synthesis, characterization and biological evaluation of different types of iron oxide nanoparticles (mNPs'), as promising material for tumor hyperthermia. The surface of mNPs' has modified with inorganic stabilizing agents to particularly improve characteristics such as their magnetic properties, colloidal stability and biocompatibility. The successful coating of mNPs' was confirmed by morphological and structural characterization by transmission electron microscopy (TEM) and Fourier-Transform Infra-Red spectroscopy (FT-IR), while their hydrodynamic diameter was studied by using Dynamic light scattering (DLS). X-ray Diffraction (XRD) proved that the crystallite phase of mNPs' is the same with the pattern of magnetite. Superparamagnetic behavior and mNPs' response under the application of alternating magnetic field (AMF) were also thoroughly investigated and showed good heating efficiency in magnetic hyperthermia experiments. The contrast ability in magnetic resonance imaging (MRI) is also discussed indicating that mNPs are negative MRI contrast types. Nonetheless the effects of mNPs on cell viability was performed by MTT on human keratinocytes, human embryonic kidney cells, endothelial cells and by hemolytic assay on erythrocytes. In healthy keratinocytes wound healing assay in different time intervals was performed, assessing both the cell migration and wound closure. Endothelial cells have also been studied in functional activity performing capillary morphogenesis. In vitro studies showed that mNPs are safely taken by the healthy cells and do not interfere with the biological processes such as cell migration and motility.

Disentangle, Align and Fuse for Multimodal and Semi-Supervised Image Segmentation.

Magnetic resonance (MR) protocols rely on several sequences to assess pathology and organ status properly. Despite advances in image analysis, we tend to treat each sequence, here termed modality, in isolation. Taking advantage of the common information shared between modalities (an organ's anatomy) is beneficial for multi-modality processing and learning. However, we must overcome inherent anatomical misregistrations and disparities in signal intensity across the modalities to obtain this benefit. We present a method that offers improved segmentation accuracy of the modality of interest (over a single input model), by learning to leverage information present in other modalities, even if few (semi-supervised) or no (unsupervised) annotations are available for this specific modality. Core to our method is learning a disentangled decomposition into anatomical and imaging factors. Shared anatomical factors from the different inputs are jointly processed and fused to extract more accurate segmentation masks. Image misregistrations are corrected with a Spatial Transformer Network, which non-linearly aligns the anatomical factors. The imaging factor captures signal intensity characteristics across different modality data and is used for image reconstruction, enabling semi-supervised learning. Temporal and slice pairing between inputs are learned dynamically. We demonstrate applications in Late Gadolinium Enhanced (LGE) and Blood Oxygenation Level Dependent (BOLD) cardiac segmentation, as well as in T2 abdominal segmentation. Code is available at https://github.com/vios-s/multimodal_segmentation.

Pharmacokinetic modelling for the simultaneous assessment of perfusion and 18F-flutemetamol uptake in cerebral amyloid angiopathy using a reduced PET-MR acquisition time: Proof of concept.

PURPOSE: Cerebral amyloid angiopathy (CAA) is a cerebral small vessel disease associated with perivascular ß-amyloid deposition. CAA is also associated with strokes due to lobar intracerebral haemorrhage (ICH). 18F-flutemetamol amyloid ligand PET may improve the early detection of CAA. We performed pharmacokinetic modelling using both full (0-30, 90-120 min) and reduced (30 min) 18F-flutemetamol PET-MR acquisitions, to investigate regional cerebral perfusion and amyloid deposition in ICH patients. METHODS: Dynamic18F-flutemetamol PET-MR was performed in a pilot cohort of sixteen ICH participants; eight lobar ICH cases with probable CAA and eight deep ICH patients. A model-based input function (mIF) method was developed for compartmental modelling. mIF 1-tissue (1-TC) and 2-tissue (2-TC) compartmental modelling, reference tissue models and standardized uptake value ratios were assessed in the setting of probable CAA detection. RESULTS: The mIF 1-TC model detected perfusion deficits and 18F-flutemetamol uptake in cases with probable CAA versus deep ICH patients, in both full and reduced PET acquisition time (all P < 0.05). In the reduced PET acquisition, mIF 1-TC modelling reached the highest sensitivity and specificity in detecting perfusion deficits (0.87, 0.77) and 18F-flutemetamol uptake (0.83, 0.71) in cases with probable CAA. Overall, 52 and 48 out of the 64 brain areas with 18F-flutemetamol-determined amyloid deposition showed reduced perfusion for 1-TC and 2-TC models, respectively. CONCLUSION: Pharmacokinetic (1-TC) modelling using a 30 min PET-MR time frame detected impaired haemodynamics and increased amyloid load in probable CAA. Perfusion deficits and amyloid burden co-existed within cases with CAA, demonstrating a distinct imaging pattern which may have merit in elucidating the pathophysiological process of CAA.

Tri-modal imaging of gold-dotted magnetic nanoparticles for magnetic resonance imaging, computed tomography and intravascular ultrasound: an in vitro study.

Aim: To examine the multimodal contrasting ability of gold-dotted magnetic nanoparticles (Au*MNPs) for magnetic resonance (MR), computed tomography (CT) and intravascular ultrasound (IVUS) imaging. Materials & methods: Au*MNPs were prepared by adapting an impregnation method, without using surface capping reagents and characterized (transmission electron microscopy, x-ray diffraction and Fourier-transform infrared spectroscopy) with their in vitro cytotoxicity assessed, followed by imaging assessments. Results: The contrast-enhancing ability of Au*MNPs was shown to be concentration-dependent across MR, CT and IVUS imaging. The Au content of the Au*MNP led to evident increases of the IVUS signal. Conclusion: We demonstrated that Au*MNPs showed concentration-dependent contrast-enhancing ability in MRI and CT imaging, and for the first-time in IVUS imaging due to the Au content. These Au*MNPs are promising toward solidifying tri-modal imaging-based theragnostics.

Bone marrow adipose tissue is a unique adipose subtype with distinct roles in glucose homeostasis.

Bone marrow adipose tissue (BMAT) comprises >10% of total adipose mass, yet unlike white or brown adipose tissues (WAT or BAT) its metabolic functions remain unclear. Herein, we address this critical gap in knowledge. Our transcriptomic analyses revealed that BMAT is distinct from WAT and BAT, with altered glucose metabolism and decreased insulin responsiveness. We therefore tested these functions in mice and humans using positron emission tomography-computed tomography (PET/CT) with 18F-fluorodeoxyglucose. This revealed that BMAT resists insulin- and cold-stimulated glucose uptake, while further in vivo studies showed that, compared to WAT, BMAT resists insulin-stimulated Akt phosphorylation. Thus, BMAT is functionally distinct from WAT and BAT. However, in humans basal glucose uptake in BMAT is greater than in axial bones or subcutaneous WAT and can be greater than that in skeletal muscle, underscoring the potential of BMAT to influence systemic glucose homeostasis. These PET/CT studies characterise BMAT function in vivo, establish new methods for BMAT analysis, and identify BMAT as a distinct, major adipose tissue subtype.

Disentangled representation learning in cardiac image analysis.

Typically, a medical image offers spatial information on the anatomy (and pathology) modulated by imaging specific characteristics. Many imaging modalities including Magnetic Resonance Imaging (MRI) and Computed Tomography (CT) can be interpreted in this way. We can venture further and consider that a medical image naturally factors into some spatial factors depicting anatomy and factors that denote the imaging characteristics. Here, we explicitly learn this decomposed (disentangled) representation of imaging data, focusing in particular on cardiac images. We propose Spatial Decomposition Network (SDNet), which factorises 2D medical images into spatial anatomical factors and non-spatial modality factors. We demonstrate that this high-level representation is ideally suited for several medical image analysis tasks, such as semi-supervised segmentation, multi-task segmentation and regression, and image-to-image synthesis. Specifically, we show that our model can match the performance of fully supervised segmentation models, using only a fraction of the labelled images. Critically, we show that our factorised representation also benefits from supervision obtained either when we use auxiliary tasks to train the model in a multi-task setting (e.g. regressing to known cardiac indices), or when aggregating multimodal data from different sources (e.g. pooling together MRI and CT data). To explore the properties of the learned factorisation, we perform latent-space arithmetic and show that we can synthesise CT from MR and vice versa, by swapping the modality factors. We also demonstrate that the factor holding image specific information can be used to predict the input modality with high accuracy. Code will be made available at https://github.com/agis85/anatomy_modality_decomposition.

Peripheral Retinal Imaging Biomarkers for Alzheimer's Disease: A Pilot Study.

PURPOSE: To examine whether ultra-widefield (UWF) retinal imaging can identify biomarkers for Alzheimer's disease (AD) and its progression. METHODS: Images were taken using a UWF scanning laser ophthalmoscope (Optos P200C AF) to determine phenotypic variations in 59 patients with AD and 48 healthy controls at baseline (BL). All living participants were invited for a follow-up (FU) after 2 years and imaged again (if still able to participate). All participants had blood taken for genotyping at BL. Images were graded for the prevalence of age-related macular degeneration-like pathologies and retinal vascular parameters. Comparison between AD patients and controls was made using the Student t test and the χ2 test. RESULTS: Analysis at BL revealed a significantly higher prevalence of a hard drusen phenotype in the periphery of AD patients (14/55; 25.4%) compared to controls (2/48; 4.2%) [χ2 = 9.9, df = 4, p = 0.04]. A markedly increased drusen number was observed at the 2-year FU in patients with AD compared to controls. There was a significant increase in venular width gradient at BL (zone C: 8.425 × 10-3 ± 2.865 × 10-3 vs. 6.375 × 10-3 ± 1.532 × 10-3, p = 0.008; entire image: 8.235 × 10-3 ± 2.839 × 10-3 vs. 6.050 × 10-3 ± 1.414 × 10-3, p = 0.004) and a significant decrease in arterial fractal dimension in AD at BL (entire image: 1.250 ± 0.086 vs. 1.304 ± 0.089, p = 0.049) with a trend for both at FU. CONCLUSIONS: UWF retinal imaging revealed a significant association between AD and peripheral hard drusen formation and changes to the vasculature beyond the posterior pole, at BL and after clinical progression over 2 years, suggesting that monitoring pathological changes in the peripheral retina might become a valuable tool in AD monitoring.

Evaluation of pharmacokinetic models for perfusion imaging with dynamic contrast-enhanced magnetic resonance imaging in porcine skeletal muscle using low-molecular-weight contrast agents.

PURPOSE: Pharmacokinetic models for perfusion quantification with a low-molecular-weight contrast agent (LMCA) in skeletal muscle using dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) were evaluated. METHODS: Tissue perfusion was measured in seven regions of interest (ROIs) placed in the total hind leg supplied by the femoral artery in seven female pigs. DCE-MRI was performed using a 3D gradient echo sequence with k-space sharing. The sequence was acquired twice, first after LMCA and then after blood pool contrast agent injection. Blood flow was augmented by continuous infusion of the vasodilator adenosine into the femoral artery, resulting in up to four times increased blood flow. The results obtained with several LMCA models were compared with those of a two-compartment blood pool model (2CBPM) consisting of a capillary and an arteriolar compartment. Measurements performed with a Doppler flow probe placed at the femoral artery served as ground truth. RESULTS: The two-compartment exchange model extended by an arteriolar compartment (E2CXM) showed the highest fit quality of all LMCA models and the most significant correlation with the Doppler measurements, r = 0.78 (P < 0.001). The best correspondence between the capillary perfusion measurements of the LMCA models and those of the 2CBPM was found with the E2CXM (slope of the regression line equal to 1, r = 0.85, P < 0.001). The results for the clinical patient data corresponded very well with the results obtained in the animal experiments. CONCLUSIONS: Double-contrast agent DCE-MRI in combination with the E2CXM yields the most reliable results and can be used in clinical routine. Magn Reson Med 79:3154-3162, 2018. © 2017 International Society for Magnetic Resonance in Medicine.

Quantitative assessment of myocardial blood flow in coronary artery disease by cardiovascular magnetic resonance: comparison of Fermi and distributed parameter modeling against invasive methods.

BACKGROUND: Mathematical modeling of perfusion cardiovascular magnetic resonance (CMR) data allows absolute quantification of myocardial blood flow and can potentially improve the diagnosis and prognostication of obstructive coronary artery disease (CAD), against the current clinical standard of visual assessments. This study compares the diagnostic performance of distributed parameter modeling (DP) against the standard Fermi model, for the detection of obstructive CAD, in per vessel against per patient analysis. METHODS: A pilot cohort of 28 subjects (24 included in the final analysis) with known or suspected CAD underwent adenosine stress-rest perfusion CMR at 3T. Data were analysed using Fermi and DP modeling against invasive coronary angiography and fractional flow reserve, acquired in all subjects. Obstructive CAD was defined as luminal stenosis of ≥70 % alone, or luminal stenosis ≥50 % and fractional flow reserve ≤0.80. RESULTS: On ROC analysis, DP modeling outperformed the standard Fermi model, in per vessel and per patient analysis. In per patient analysis, DP modeling-derived myocardial blood flow at stress demonstrated the highest sensitivity and specificity (0.96, 0.92) in detecting obstructive CAD, against Fermi modeling (0.78, 0.88) and visual assessments (0.79, 0.88), respectively. CONCLUSIONS: DP modeling demonstrated consistently increased diagnostic performance against Fermi modeling and showed that it may have merit for stratifying patients with at least one vessel with obstructive CAD. CLINICAL TRIAL REGISTRATION: Clinicaltrials.gov NCT01368237 Registered 6 of June 2011. URL: https://clinicaltrials.gov/ct2/show/NCT01368237.

Measurement of myocardial blood flow by cardiovascular magnetic resonance perfusion: comparison of distributed parameter and Fermi models with single and dual bolus.

BACKGROUND: Mathematical modeling of cardiovascular magnetic resonance perfusion data allows absolute quantification of myocardial blood flow. Saturation of left ventricle signal during standard contrast administration can compromise the input function used when applying these models. This saturation effect is evident during application of standard Fermi models in single bolus perfusion data. Dual bolus injection protocols have been suggested to eliminate saturation but are much less practical in the clinical setting. The distributed parameter model can also be used for absolute quantification but has not been applied in patients with coronary artery disease. We assessed whether distributed parameter modeling might be less dependent on arterial input function saturation than Fermi modeling in healthy volunteers. We validated the accuracy of each model in detecting reduced myocardial blood flow in stenotic vessels versus gold-standard invasive methods. METHODS: Eight healthy subjects were scanned using a dual bolus cardiac perfusion protocol at 3T. We performed both single and dual bolus analysis of these data using the distributed parameter and Fermi models. For the dual bolus analysis, a scaled pre-bolus arterial input function was used. In single bolus analysis, the arterial input function was extracted from the main bolus. We also performed analysis using both models of single bolus data obtained from five patients with coronary artery disease and findings were compared against independent invasive coronary angiography and fractional flow reserve. Statistical significance was defined as two-sided P value < 0.05. RESULTS: Fermi models overestimated myocardial blood flow in healthy volunteers due to arterial input function saturation in single bolus analysis compared to dual bolus analysis (P < 0.05). No difference was observed in these volunteers when applying distributed parameter-myocardial blood flow between single and dual bolus analysis. In patients, distributed parameter modeling was able to detect reduced myocardial blood flow at stress (<2.5 mL/min/mL of tissue) in all 12 stenotic vessels compared to only 9 for Fermi modeling. CONCLUSIONS: Comparison of single bolus versus dual bolus values suggests that distributed parameter modeling is less dependent on arterial input function saturation than Fermi modeling. Distributed parameter modeling showed excellent accuracy in detecting reduced myocardial blood flow in all stenotic vessels.