Searcharxiv⌕ Search

arXiv subjects

Kevin M. Koch

Publications and source records attributed to Kevin M. Koch.

15 recordsLinked to original sources
arXiv2026

Measurements Automatically Extracted from Zero Echo Time MRI Using Deep Learning Image Segmentation and Geometric Modeling Agree with Expert Manual Readings

Computed tomography (CT) remains the reference for 3D osseous morphometry in femoroacetabular impingement (FAI) but requires ionizing radiation and manual measurement. Zero echo time (ZTE) MRI visualizes cortical bone and yields FAI angles that agree with CT, but automated angle extraction remains limited. We developed and validated automated FAI angle computation from ZTE MRI and assessed agreement with expert manual measurements in a cross-sectional study (level of evidence, 3). Pelvic ZTE MRI was acquired in 73 participants (mean age 36.8 +/- 18.5 years; 51 women, 22 men), yielding 135 hips. nnU-Net was trained on 100 manually curated hips to segment the femur, pelvis, and three osseous landmarks. Custom geometric algorithms computed alpha, femoral neck-shaft, Tonnis, coronal and sagittal center-edge, and acetabular version angles from inferred segmentations. Measurements on 35 test hips were compared with the mean of two radiologists' manual measures using intraclass correlation (ICC) and Bland-Altman analysis. Dice exceeded 0.96 for bone and ranged from 0.65 to 0.83 for landmarks. Median landmark error was 0.38 mm (femoral head), 0.82 mm (lateral acetabulum), and <2.5 mm (medial acetabulum, greater trochanter). Interrater ICC was excellent for acetabular version, coronal center-edge, and Tonnis (>=0.82) but poor for alpha and femoral neck-shaft. Model versus rater-mean agreement was excellent for acetabular version, coronal center-edge, and Tonnis (0.92-0.96), good for mid-acetabular sagittal center-edge (0.74), and fair for alpha (0.45) and femoral neck-shaft (0.55). Model Bland-Altman limits of agreement were narrower than interrater limits for most angles. Fully automated morphometric assessment from ZTE MRI is feasible and performs comparably to expert readers for most coverage and version angles.

physics.med-ph↗
arXiv2024

Variable Resolution Sampling and Deep Learning Image Recovery for Accelerated Multi-Spectral MRI Near Metal Implants

Purpose: This study presents a variable resolution (VR) sampling and deep learning reconstruction approach for multi-spectral MRI near metal implants, aiming to reduce scan times while maintaining image quality. Background: The rising use of metal implants has increased MRI scans affected by metal artifacts. Multi-spectral imaging (MSI) reduces these artifacts but sacrifices acquisition efficiency. Methods: This retrospective study on 1.5T MSI knee and hip data from patients with metal hardware used a novel spectral undersampling scheme to improve acquisition efficiency by ~40%. U-Net-based deep learning models were trained for reconstruction. Image quality was evaluated using SSIM, PSNR, and RESI metrics. Results: Deep learning reconstructions of undersampled VR data (DL-VR) showed significantly higher SSIM and PSNR values (p<0.001) compared to conventional reconstruction (CR-VR), with improved edge sharpness. Edge sharpness in DL-reconstructed images matched fully sampled references (p=0.5). Conclusion: This approach can potentially enhance MRI examinations near metal implants by reducing scan times or enabling higher resolution. Further prospective studies are needed to assess clinical value.

eess.IV↗
arXiv2023

On Functional Activations in Deep Neural Networks

Background: Deep neural networks have proven to be powerful computational tools for modeling, prediction, and generation. However, the workings of these models have generally been opaque. Recent work has shown that the performance of some models are modulated by overlapping functional networks of connections within the models. Here the techniques of functional neuroimaging are applied to an exemplary large language model to probe its functional structure. Methods: A series of block-designed task-based prompt sequences were generated to probe the Facebook Galactica-125M model. Tasks included prompts relating to political science, medical imaging, paleontology, archeology, pathology, and random strings presented in an off/on/off pattern with prompts about other random topics. For the generation of each output token, all layer output values were saved to create an effective time series. General linear models were fit to the data to identify layer output values which were active with the tasks. Results: Distinct, overlapping networks were identified with each task. Most overlap was observed between medical imaging and pathology networks. These networks were repeatable across repeated performance of related tasks, and correspondence of identified functional networks and activation in tasks not used to define the functional networks was shown to accurately identify the presented task. Conclusion: The techniques of functional neuroimaging can be applied to deep neural networks as a means to probe their workings. Identified functional networks hold the potential for use in model alignment, modulation of model output, and identifying weights to target in fine-tuning.

cs.AI↗
arXiv2023

Variable Resolution Sampling and Deep Learning-Based Image Recovery for Faster Multi-Spectral Imaging Near Metal Implants

Purpose: In multi-spectral imaging (MSI), several fast spin echo volumes with discrete Larmor frequency offsets are acquired in an interleaved fashion with multiple concatenations. Here, a variable resolution (VR) method to nearly halve scan time is proposed by only acquiring low resolution autocalibrating signal in half of the concatenations. Methods: Knee MSI datasets were retrospectively undersampled with the proposed variable resolution sampling scheme. A U-Net model was trained to predict the full-resolution images from the VR input. Image quality was assessed in 10 test subjects. Results: Spectral bin-combined images produced with the proposed variable resolution sampling with deep learning reconstruction appear to be of high quality and exhibited a median structural image similarity of 0.984 across test subjects and slices. Conclusion: The proposed variable resolution sampling method shows promise for drastically reducing the time it takes to collect multi-spectral imaging data near metallic implants. Further studies will rigorously examine its clinical utility across multiple implant scenarios.

physics.med-ph↗
arXiv2021

Unconstrained Kinematic MRI Tracking of Wrist Carpal Bones

In this preliminary study technical methodology for kinematic tracking and profiling of wrist carpal bones during unconstrained movements is explored. Heavily under-sampled and fat-saturated 3D Cartesian MRI acquisition were used to capture temporal frames of the unconstrained moving wrist of 5 healthy subjects. A slab-to-volume point-cloud based registration was then utilized to register the moving volumes to a high-resolution image volume set collected at a neutral resting position. Comprehensive error analyses for different acquisition parameter settings were performed to evaluate the performance limits of several derived kinematic metrics. Computational results suggested that sufficient volume coverage for the dynamic acquisitions was reached when collecting 12 slice-encodes at 2.5mm resolution, which yielded a temporal resolution of and 2.57 seconds per volumetric frame. These acquisition parameters resulted in total absolute errors of 1.9$^\circ\pm$1.8$^\circ$ (3$^\circ\pm$4.6$^\circ$) in derived rotation angles and 0.3mm$\pm$0.47mm (0.72mm$\pm$0.8mm) in center-of-mass displacement kinematic profiles within ulnar-radial (flexion-extension) motion. The results of this study have established the feasibility of kinematic metric tracking of unconstrained wrist motion using 4D MRI. Temporal metric profiles derived from ulnar-radial deviation motion demonstrated better performance than those derived from flexion/extension movements. Future work will continue to explore the use of these methods in deriving more complex kinematic metrics and their application to subjects with symptomatic carpal dysfunction.

physics.med-ph↗
arXiv2021

Deep-learning based Tools for Automated Protocol Definition of Advanced Diagnostic Imaging Exams

Purpose: This study evaluates the effectiveness and impact of automated order-based protocol assignment for magnetic resonance imaging (MRI) exams using natural language processing (NLP) and deep learning (DL). Methods: NLP tools were applied to retrospectively process orders from over 116,000 MRI exams with 200 unique sub-specialized protocols ("Local" protocol class). Separate DL models were trained on 70\% of the processed data for "Local" protocols as well as 93 American College of Radiology ("ACR") protocols and 48 "General" protocols. The DL Models were assessed in an "auto-protocoling (AP)" inference mode which returns the top recommendation and in a "clinical decision support (CDS)" inference mode which returns up to 10 protocols for radiologist review. The accuracy of each protocol recommendation was computed and analyzed based on the difference between the normalized output score of the corresponding neural net for the top two recommendations. Results: The top predicted protocol in AP mode was correct for 82.8%, 73.8%, and 69.3% of the test cases for "General", "ACR", and "Local" protocol classes, respectively. Higher levels of accuracy over 96% were obtained for all protocol classes in CDS mode. However, at current validation performance levels, the proposed models offer modest, positive, financial impact on large-scale imaging networks. Conclusions: DL-based protocol automation is feasible and can be tuned to route substantial fractions of exams for auto-protocoling, with higher accuracy with more general protocols. Economic analyses of the tested algorithms indicate that improved algorithm performance is required to yield a practical exam auto-protocoling tool for sub-specialized imaging exams.

cs.LG↗
arXiv2021

Diffusion propagator metrics are biased when simultaneous multi-slice acceleration is used

Advanced diffusion MRI models are being explored to study the complex microstructure of the brain with higher accuracy. However, these techniques require long acquisition times. Simultaneous multi-slice (SMS) accelerates data acquisition by exciting multiple image slices simultaneously and separating the overlapping slices using a mathematical model. However, this slice separation is not exact and leads to crosstalk between simultaneously excited slices. Although this residual leakage is small, it affects quantitative MRI techniques such as diffusion imaging. In this study, the effects of SMS acceleration on the accuracy of propagator metrics obtained from the MAP-MRI technique was investigated. Ten healthy volunteers were scanned with SMS accelerated multi-shell diffusion MRI acquisitions. Group analyses were performed to study brain regions typically affected by SMS acceleration. In addition, diffusion metrics from atlas based fiber tracts of interest were analyzed to investigate how propagator metrics in major fiber tracts were biased by 2- and 3-band SMS acceleration. Both zero-displacement metrics and non-Gaussianity metrics were significantly altered when SMS acceleration was used. MAP-MRI metrics calculated from SMS-3 showed significant differences with respect to SMS-2. Furthermore, when shorter TR afforded by SMS acceleration was used, the characteristics of this bias have changed. This has implications for studies using diffusion MRI with SMS acceleration to investigate the effects of a disease or injury on the brain tissues.

physics.med-ph↗
arXiv2020

Model-based Learning for Quantitative Susceptibility Mapping

Quantitative susceptibility mapping (QSM) is a magnetic resonance imaging (MRI) technique that estimates magnetic susceptibility of tissue from Larmor frequency offset measurements. The generation of QSM requires solving a challenging ill-posed field-to-source inversion problem. Inaccurate field-to-source inversion often causes large susceptibility estimation errors that appear as streaking artifacts in the QSM, especially in massive hemorrhagic regions. Recently, several deep learning (DL) QSM techniques have been proposed and demonstrated impressive performance. Due to the inherent non-existent ground-truth QSM references, these DL techniques used either calculation of susceptibility through multiple orientation sampling (COSMOS) maps or synthetic data for network training. Therefore, they were constrained by the availability and accuracy of COSMOS maps, or suffered from performance drop when the training and testing domains were different. To address these limitations, we present a model-based DL method, denoted as uQSM. Without accessing to QSM labels, uQSM is trained using the well-established physical model. When evaluating on multi-orientation QSM datasets, uQSM achieves higher levels of quantitative accuracy compared to TKD, TV-FANSI, MEDI, and DIP approaches. When qualitatively evaluated on single-orientation datasets, uQSM outperforms other methods and reconstructed high quality QSM.

physics.med-ph↗
arXiv2019

Meta-QSM: An Image-Resolution-Arbitrary Network for QSM Reconstruction

Quantitative Susceptibility Mapping (QSM) can estimate the underlying tissue magnetic susceptibility and reveal pathology. Current deep-learning-based approaches to solve the QSM inverse problem are restricted on fixed image resolution. They trained a specific model for each image resolution which is inefficient in computing. In this work, we proposed a novel method called Meta-QSM to firstly solve QSM reconstruction of arbitrary image resolution with a single model. In Meta-QSM, weight prediction was used to predict the weights of kernels by taking the image resolution as input. The proposed method was evaluated on synthetic data and clinical data with comparison to existing QSM reconstruction methods. The experimental results showed the Meta-QSM can effectively reconstruct susceptibility maps with different image resolution using one neural network training.

physics.med-ph↗
arXiv2019

Deep Quantitative Susceptibility Mapping for Background Field Removal and Total Field Inversion

Quantitative susceptibility mapping (QSM) utilizes MRI signal phase to estimate local tissue susceptibility, which has been shown useful to provide novel image contrast and as biomarkers of abnormal tissue. QSM requires addressing a challenging post-processing problem: filtering of image phase estimates and inversion of the phase to susceptibility relationship. A wide variety of quantification errors, robustness limitations, and artifacts constraints QSM clinical translation. To overcome these limitations, a robust deep-learning-based QSM reconstruction approach is proposed to perform background field removal and susceptibility inversion simultaneously from input MRI phase images. Synthetic training data based on in-vivo data sources and physics simulations were used for training. The network was quantitatively tested using gold-standard in-silico labeled dataset against established background field removal and QSM inversion approaches. In addition, the algorithm was applied to a QSM challenge data and clinical susceptibility-weighted imaging (SWI) data. When quantitatively compared against gold-standard in-silico labels, the proposed algorithm outperformed the existing comparable background field removal approaches and QSM reconstruction algorithms. The QSM challenge data and clinical SWI data demonstrated that the proposed approach was able to robustly generate high quality local field and QSM with improved accuracy.

physics.med-ph↗
arXiv2019

Non-locally Encoder-Decoder Convolutional Network for Whole Brain QSM Inversion

Quantitative Susceptibility Mapping (QSM) reconstruction is a challenging inverse problem driven by ill conditioning of its field-to -susceptibility transformation. State-of-art QSM reconstruction methods either suffer from image artifacts or long computation times, which limits QSM clinical translation efforts. To overcome these limitations, a non-locally encoder-decoder gated convolutional neural network is trained to infer whole brain susceptibility map, using the local field and brain mask as the inputs. The performance of the proposed method is evaluated relative to synthetic data, a publicly available challenge dataset, and clinical datasets. The proposed approach can outperform existing methods on quantitative metrics and visual assessment of image sharpness and streaking artifacts. The estimated susceptibility maps can preserve conspicuity of fine features and suppress streaking artifacts. The demonstrated methods have potential value in advancing QSM clinical research and aiding in the translation of QSM to clinical operations.

cs.AI↗
arXiv2019

MRI Tissue Magnetism Quantification through Total Field Inversion with Deep Neural Networks

Quantitative susceptibility mapping (QSM) utilizes MRI signal phase to infer estimates of local tissue magnetism (magnetic susceptibility), which has been shown useful to provide novel image contrast and as biomarkers of abnormal tissue. QSM requires addressing a challenging post-processing problem: filtering of image phase estimates and inversion of the phase to susceptibility relationship. A wide variety of quantification errors, robustness limitations, and artifacts plague QSM algorithms. To overcome these limitations, a robust deep-learning-based single-step QSM reconstruction approach is proposed and demonstrated. This neural network was trained using magnetostatic physics simulations based on in-vivo data sources. Random perturbations were added to the physics simulations to provide sufficient input-label pairs for the training purposes. The network was quantitatively tested using gold-standard in-silico labeled datasets against established QSM total field inversion approaches. In addition, the algorithm was applied to susceptibility-weighted imaging (SWI) data collected on a cohort of clinical subjects with brain hemmhorage. When quantitatively compared against gold-standard in-silico labels, the proposed algorithm outperformed the existing comparable approaches. High quality QSM were consistently estimated from clinical susceptibility-weighted data on 100 subjects without any noticeable inversion failures. The proposed approach was able to robustly generate high quality QSM with improved accuracy in in-silico gold-standard experiments. QSM produced by the proposed method can be generated in real-time on existing MRI scanner platforms and provide enhanced visualization and quantification of magnetism-based tissue contrasts.

physics.med-ph↗
arXiv2019

Quantitative Susceptibility Inversion Through Parcellated Multiresolution Neural Networks and K-Space Substitution

Purpose: Quantitative Susceptibility Mapping (QSM) reconstruction is a challenging inverse problem driven by poor conditioning of the field to susceptibility transformation. State-of-art QSM reconstruction methods either suffer from image artifacts or long computation times, which limits QSM clinical translation efforts. To overcome these limitations, a deep-learning-based approach is proposed and demonstrated. Methods: An encoder-decoder neural network was trained to infer susceptibility maps on volume parcellated regions. The training data consisted of fabricated susceptibility distributions modeled to mimic the spatial frequency patterns of in-vivo brain susceptibility distributions. Inferred volume parcels were recombined to form composite QSM. This approach is denoted as ASPEN, standing for Approximated Susceptibility through Parcellated Encoder-decoder Networks. ASPEN performance was evaluated relative to several well-established approaches on a gold-standard challenge dataset and on cohort of 200 study subjects. Results: ASPEN provided similar levels of quantitative accuracy compared to the evaluated established approaches on the gold standard ISMRM Challenge dataset, but qualitatively showed marked reductions in streaking artifacts and map blurring. On the large-cohort dataset, ASPEN achieved the highest score compared with other methods in a multi-rater evaluation of streaking artifacts and map resolution. Conclusion: The proposed ASPEN approach can robustly infer susceptibility maps in near real-time on routine computational hardware. This preliminary study establishes ASPEN's parity with existing approaches for quantitative accuracy on a well-curated gold standard dataset and further demonstrates its robustness to streaking artifacts across a large cohort of subjects.

physics.med-ph↗
arXiv2019

Build-A-FLAIR: Synthetic T2-FLAIR Contrast Generation through Physics Informed Deep Learning

Purpose: Magnetic resonance imaging (MRI) exams include multiple series with varying contrast and redundant information. For instance, T2-FLAIR contrast is based upon tissue T2 decay and the presence of water, also present in T2- and diffusion-weighted contrasts. T2-FLAIR contrast can be hypothetically modeled through deep learning models trained with diffusion- and T2-weighted acquisitions. Methods: Diffusion-, T2-, T2-FLAIR-, and T1-weighted brain images were acquired in 15 individuals. A convolutional neural network was developed to generate a T2-FLAIR image from other contrasts. Two datasets were withheld from training for validation. Results: Inputs with physical relationships to T2-FLAIR contrast most significantly impacted performance. The best model yielded results similar to acquired T2-FLAIR images, with a structural similarity index of 0.909, and reproduced pathology excluded from training. Synthetic images qualitatively exhibited lower noise and increased smoothness compared to acquired images. Conclusion: This suggests that with optimal inputs, deep learning based contrast generation performs well with creating synthetic T2-FLAIR images. Feature engineering on neural network inputs, based upon the physical basis of contrast, impacts the generation of synthetic contrast images. A larger, prospective clinical study is needed.

physics.med-ph↗