Quantitative estimation of biological tissue parameters using magnetic resonance

Abstract

A method of optimizing a magnetic resonance (MR) pulse sequence for generating quantitative maps of biological parameters. The method includes receiving, by a processor, quantitative magnetic resonance (qMR) parameters relating to a tissue of interest. The method includes generating, by one or more machine learning models based on the qMR parameters and an optimization objective, a tailored pulse sequence sensitive to biological parameters associated with the tissue of interest.

Description

QUANTITATIVE ESTIMATION OF BIOLOGICAL TISSUE PARAMETERS USING MAGNETIC RESONANCECROSS-REFERENCE TO RELATED PATENT APPLICATION

[0001] This application claims the benefit and priority of U.S. Provisional Patent App. No. 63 / 522,653, filed on June 22, 2023, the entirety of which is incorporated by reference herein.STATEMENT OF GOVERNMENT INTEREST

[0002] This invention was made with government support under contract nos. R01 NS 131948,R01 AR070297, R21 EB020096, P41 EB017183, and R01 NS131948, awarded by the National Institutes of Health. The government has certain rights in the invention.TECHNICAL FIELD

[0003] The present disclosure relates generally to systems and methods for quantitative estimation of biological tissue parameters using magnetic resonance imaging (“MRI”).BACKGROUND

[0004] MRI is a medical imaging technique used to acquire images of patients and is routinely used to diagnose disease, guide therapy, and / or to monitor the progress of disease in patients. Quantitative magnetic resonance (“qMR”) techniques provide quantitative measurements of physical and physiological parameters within tissues, rather than anatomical images, and are used to extract quantitative biological markers (e.g., biomarkers, biological parameters) from MR data.

[0005] Model-based qMR techniques rely on matching experimental MR data to simulated MR data to estimate tissue-specific parameters. These qMR parameters (e.g., T1 relaxation time, T2 relaxation time, etc.) are typically used to obtain insight into the underlying biology of the tissue. While the qMR parameters are correlated with tissue biological features, they do not reflect an individual tissue component or biological parameter. The use of biological markers in mapping can instead enable accurate tracking of biological changes longitudinally, drug and / or treatment efficacy, and disease progression. Therefore, the ability to map biological markers is useful in diagnosis and treatment, as a clinical tool, and as a research tool.

[0006] qMR encompasses a number of techniques, such as T1 mapping, T2 mapping, Tip mapping, magnetization transfer (“MT”), diffusion weighted imaging (“DWI”), diffusion tenorimaging (“DTI”), etc. Magnetic Resonance Fingerprinting (“MRF”) is an example of a modelbased qMR technique for quantitative estimation of tissue-specific parameters from MR data. MRF techniques have been developed to allow for the estimation of multiple qMR parameters with a single MR acquisition. MRF enables spatial quantification (mapping) of multiple physical parameters in an accurate, reliable, and reproducible way within the constraints of clinically feasible MR scan time.

[0007] MRF addresses some of the main technological challenges that have traditionally prevented the clinical translation of quantitative MRI. However, MRF, like other qMR techniques, is rarely specific to individual biological changes, hampering interpretation. In contrast, histology provides specific information with a straightforward interpretation. However, histology is an invasive procedure that requires ex vivo tissue specimens.SUMMARY

[0008] At least one embodiment relates to a system for generating quantitative maps of biological parameters including a processor and a memory coupled to the processor. The memory has instructions stored thereon which, when executed by the processor, cause the system to generate, using one or more machine learning models, a tailored pulse sequence based on one or more biological parameters associated with a tissue of interest. The system sends the tailored pulse sequence to a magnetic resonance (“MR”) scanner and receives MR signals for the tissue of interest. The MR signals are associated with the tailored pulse sequence. The system generates, using the one or more machine learning models, quantitative histology maps from the MR signal.

[0009] One embodiment relates to a non-transitory processor-readable medium storing code representing instructions to be executed by one or more processors. The instructions include code to cause the one or more processors to generate, using one or more machine learning models, a tailored pulse sequence sensitive to one or more biological parameters associated with a tissue of interest. The one or more machine learning models is trained to optimize the tailored pulse sequence based on an optimization objective. The instructions include code to cause the one or more processors to send the tailored pulse sequence to an MR scanner and receive MR signals for the tissue of interest. The MR signals are associated with the tailored pulse sequence. The instructions include code to cause the one or more processors to generate, using the one or more machine learning models, quantitative histology maps from the MR signals.

[0010] One embodiment relates to a method of optimizing a magnetic resonance (“MR”) pulse sequence for generating quantitative maps of biological parameters. The method includes receiving, by a processor, quantitative magnetic resonance (“qMR”) parameters relating to a tissue of interest. The method includes generating, by one or more machine learning models based on the qMR parameters and an optimization objective, a tailored pulse sequence sensitive to biological parameters associated with the tissue of interest.

[0011] This summary is illustrative only and is not intended to be in any way limiting.BRIEF DESCRIPTION OF THE DRAWINGS

[0012] A conception of the advantages and features constituting the present disclosure, and of the construction and operation of typical mechanisms provided with the present disclosure, will become more readily apparent by referring to the exemplary, and therefore non-limiting, embodiments illustrated in the drawings accompanying and forming a part of this specification, wherein like reference numerals designate the same elements in the several views.

[0013] FIG. 1 shows a schematic illustration of a system for generating quantitative maps of biological parameters, according to one embodiment.

[0014] FIG. 2 shows a method of generating quantitative histology maps from MR signals, according to one embodiment.

[0015] FIG. 3 shows a method of generating quantitative maps of biological parameters, according to one embodiment.

[0016] FIG. 4 shows MR relaxation times (adiabatic Tirho, CW-Tlrho for various spin-lock frequencies, adiabatic T2A0, TRAFF, Tisat, DE-T2, T2, Ti, Tied) at 9.4 T next to histology (Safranin- O stained microscopy, collagen anisotropy (parallelism index, PI) map, amide I and carbohydrate region absorbance maps).

[0017] FIG. 5 shows depth-wise profiles of proteoglycan content and collagen fiber orientation, measured ex vivo with digital densitometry (“DD”) and polarized light microscopy (“PLM”), respectively, successfully predicted from multiple qMR data.

[0018] FIG. 6 shows preliminary data for predicting proteoglycan content and collagen network from in vivo 3 T MRF of the knee (unpublished).

[0019] FIG. 7 shows a 2D pulse sequence (“IRFF”) based on the principles of MRF for rapid qMR of cartilage.

[0020] FIG. 8 shows a modular structure of the IRFF pulse sequence.

[0021] FIG. 9 shows a sketch of the network architecture to design (“an MRI”) MRF pulse sequence optimized for the quantification of histological parameters.

[0022] The foregoing and other features of the present disclosure will become apparent from the following description and appended claims, taken in conjunction with the accompanying drawings. Understanding that these drawings depict only several embodiments in accordance with the disclosure and are therefore, not to be considered limiting in its scope, the disclosure will be described with additional specificity and detail through the use of the accompanying drawings.DETAILED DESCRIPTION

[0023] In the following detailed description, reference is made to the accompanying drawings, which form a part hereof. In the drawings, similar symbols typically identify similar components, unless context dictates otherwise. The illustrative embodiments described in the detailed description, drawings, and claims are not meant to be limiting. Other embodiments may be utilized, and other changes may be made, without departing from the spirit or scope of the subject matter presented here. It will be readily understood that the aspects of the present disclosure, as generally described herein, and illustrated in the figures, can be arranged, substituted, combined, and designed in a wide variety of different configurations, all of which are explicitly contemplated and made part of this disclosure.

[0024] Implementations herein relate to systems and methods for the quantitative estimation of biological tissue parameters using MR. The techniques of embodiments described herein utilize deep learning to generate synthetic quantitative histology maps from comprehensive information embedded in MR signals, based on information obtained with qMR techniques. A quantitative histology map provides spatial representation or visualization of quantitative data derived from histological analysis of tissue samples, A machine learning model is trained to generate quantitative histology maps (e.g., cellular density maps, morphometric maps, staining intensity maps, fibrosis quantification maps, vascular density and perfusion maps, bone microstructure maps, inflammatory cell distribution maps, neuronal connectivity maps, etc.) of biologically relevant parameters from multiple qMR parameters. For example, for articular cartilage, the machine learning model is trained to take maps of multiple qMR parameters that encodeinformation about cartilage composition and structure and generate synthetic histology maps of proteoglycan content and collagen.

[0025] Further, the techniques of embodiments described herein involve embedding the machine learning model into MR pulse sequence design to tailor MR pulse sequences for improved sensitivity to biological parameters, rather than MR parameters. As such, the MR pulse sequence can be used to tailor MRI acquisition specifically to generate quantitative histology maps. For example, systems and methods in accordance with the present disclosure can tailor the MR pulse sequence by selecting one or more parameters (e.g., flip angles, pulse durations, bandwidth, field of view, echo train length, echo time, repetition time, matrix size, etc.) such that the MR pulse sequences have greater sensitivity to one or more targeted biological parameters (e.g., tissue-specific biological properties, such as myelin and iron content in brain tissue, genetic signatures of brain tumors, proteoglycans and collagen content in cartilage, etc.) , as compared with MR pulse sequences that may be configured for sensitivity to MR parameters. The trained machine learning model can, for example, be integrated into an optimization framework to design pulse sequences. Pulse sequences design may be optimized to maximize sensitivity to biological parameters, or constituents of various tissues. Utilizing pulse sequences tailored directly for the estimation of histology parameters rather than for the estimation of MR parameters improves efficiency and sensitivity. The tailored pulse sequences are paired with a reconstruction pipeline using the machine learning model to directly reconstruct and generate accurate and reliable quantitative histology maps of biological parameters of interest directly from MR signals.

[0026] For example, the systems and methods described herein can be used in monitoring the progress of disease in patients, such as osteoarthritis (“OA”). OA is the most prevalent joint disease and a leading source of chronic pain and disability in the United States. Knee OA, in particular, accounts for more than 80% of the disease’s total burden. OA has no cure, and pathology develops from healthy to symptomatic early-stage OA, established OA, and finally, end-stage OA. Changes at early stages include hydration, proteoglycan loss, and thinning and disruption of collagen. Changes in late stages include dehydration, extensive fibrillation, loss of cartilage, and eventually, denudation of the subchondral bone. Therefore, as OA progression is a continuum, early-stage detection, and staging are important for proper clinical management and research.

[0027] MRI enables the evaluation of structural changes during the progression of knee OA, as it allows the visualization of all the tissues involved in OA pathology, such as cartilage, subchondral bone, meniscus, and soft tissue. However, for cartilage, standard MRI can onlyidentify macroscopic changes, and cannot show biochemical and microstructural changes at early stages of OA. To address this, qMR techniques have been developed to assess cartilage structure and composition. As most qMR parameters are not specific to changes in individual cartilage properties, diagnostic and prognostic interpretation is hampered. As a result, histology remains the most important outcome tool to substantiate the severity of cartilage pathology, but it can be performed only on ex-vivo samples following surgery.

[0028] In some embodiments, the techniques described herein enable noninvasive histologic analysis of cartilage using MR. Noninvasive histologic analysis of cartilage is particularly beneficial for early OA diagnosis, as well as for monitoring the effectiveness of treatment and / or cartilage repair procedures. Further, as knee OA is heterogeneous in terms of risk factors and rates of progression, the resulting virtual histology allows healthcare providers to phenotype and stratify patients to personalize treatments and improve outcomes. Additionally, noninvasive monitoring of the histological quality of cartilage longitudinally may be employed, which existing approaches do not provide. In some embodiments, the techniques described herein provide new tools to study OA onset and progression.

[0029] While certain embodiments described herein are described in reference to knee OA, the systems and methods described herein can be used to analyze other joints, for example, to detect early degenerative changes in the hip cartilage of patients with femoroacetabular impingement. Further, it can be used to analyze other biological tissues and properties. For example, the systems and methods may be applied to myelin, iron content, and / or inflammatory markers in brain tissue, or to the molecular biology measurements and genetic signatures of brain tumors. Further, it can be used to assess the efficacy of disease modifying drugs noninvasively.

[0030] Referring to FIG. 1 , a schematic illustration of a system 100 for generating quantitative maps of biological parameters is shown. Within the system 100, a pulse sequence 110 is executed by a MR Scanner 120 and generates MR signals 130, from which qMR parameters 140 are estimated. A machine learning model 150 is used to estimate one or more quantitative histology parameters either from the qMR parameters 140 or directly from the MR signals 130. In some embodiments, the machine learning model 150 includes one or more machine learning models. The machine learning model 150 can be configured (e.g., trained, updated, fine-tuned) based on training data that includes ground truth histology parameters, estimated histology parameters corresponding to ground truth histology parameters, example and / or simulated MR signals and / or qMR parameters associated with the corresponding ground truth histology parameters. Theoptimized pulse sequence design propagates information back from the quantitative histology estimates to the pulse sequence.

[0031] In some embodiments, the system 100 can include a memory and / or a processor. The processor is configured to execute instructions that are stored in the memory. For example, the machine learning model 150 can be stored in the memory and the processor can be configured to execute instructions associated with the machine learning model 150. The machine learning model 150 is trained using quantitative histological parameters. The machine learning model 150 generates synthetic quantitative histology maps from MR measurements, thereby providing “virtual histology.” The histology map can include histological images with quantitative data (e.g., cell counts, dimensions, densities, etc.) obtained through measurements or analyses of specific biological parameters within a tissue. The quantitative data is spatially mapped onto the histological images or tissue sections to provide a visual representation of the quantitative information.

[0032] The instructions, when executed by the processor can cause the system 100 generate a tailored pulse sequence 170 (e.g., MR pulse sequence, qMR pulse sequence). The tailored pulse sequence 170 is based on one or more biological parameters associated with a tissue of interest. Further, the system 100 may receive MR signals 130 associated with the tailored pulse sequence 170. In some embodiments, the system 100 receives qMR parameters 140 associated with the MR signals 130. The system 100 may generate, using the machine learning model 150, quantitative histology maps 160 from the MR signals 130.

[0033] In some embodiments, the machine learning model 150 is incorporated in an optimization framework that can be executed by the system 100 to design pulse sequences that are sensitive to tissue-specific biological tissues and properties. Typical pulse sequences are optimized to be sensitive to MR-specific physical properties (e.g., relaxation times). The optimization framework utilizes the machine learning model 150 to design the tailored pulse sequence 170. For example. A neural network can be incorporated into the optimization framework to design the pulse sequence design to be optimized to be sensitive to tissue-specific biological properties that can currently only be measured with invasive modalities such as histology. The optimization framework tailors the tailored pulse sequence 170 to the estimation of biological properties, such as quantitative histological parameters, thereby improving the sensitivity of the MR signal 130 to the biological properties of the tissue of interest.

[0034] In some embodiments, the system 100 trains the machine learning model 150 based on MR data relating to a tissue of interest as well as qMR maps and quantitative histological data associated with the MR data. This can include, for example, training the machine learning model 150 using real and / or simulated data, such as to perform a first training of one or more first layers of the machine learning model 150 using simulated data, and performing a second training (e.g., fine-tuning, such as by freezing the one or more first layers) of one or more second layers of the machine learning model 150 using real data, or vice versa. In some embodiments, the machine learning model 150 is trained to map from qMR parameters 140 to histology parameters. In some embodiments, the system 100 trains the machine learning model 150 based on the generated quantitative histology maps 160, such as based on feedback regarding the generated quantitative histology maps 160. In some embodiments, the system 100 identifies, using the machine learning model 150, a combination of qMR parameters 140 associated with selected biological parameters associated with the tissue of interest.

[0035] In some embodiments, the machine learning model 150 may include deep learning algorithms and / or other artificial intelligence algorithms. In some embodiments, the machine learning model 150 may include a multilayer perceptron (“MLP”) neural network and / or multivariate regression methods. In some embodiments, the machine learning model 150 may include Gaussian process-based methods. In some embodiments, the machine learning model 150 may include a U-Net. Once it is trained, the machine learning model 150 receives qMR parameters, such as extracted from MR data from an MRI machine, and generates a synthetic quantitative histology map 160.

[0036] Referring to FIG. 2, a method 200 of generating quantitative histology maps from MR signals is shown. In 210, a tailored pulse sequence is generated using a machine learning model. The tailored pulse sequence is sensitive to one or more biological parameters associated with a tissue of interest. The machine learning model is trained to optimize the tailored pulse sequence based on an optimization objective. In 220, the tailored pulse sequence is sent to an MR scanner to execute. In 230, MR signals based on the tailored pulse sequence are received. In 240, using the machine learning model, quantitative histology maps are generated from the MR signals.

[0037] Referring to FIG. 3, a method 300 of optimizing a MR pulse sequence for generating quantitative maps of biological parameters is shown. In 310, qMR parameters relating to a tissue of interest are received. In 320, based on the qMR parameters and an optimization objective, the machine learning model generates a tailored pulse sequence sensitive to biological parameters associated with the tissue of interest. In some embodiments the optimization objective includesincreasing sensitivity to the biological parameters and / or a threshold value for an expected sensitivity to the biological parameters. In some embodiments, the method 300 includes receiving MR signals generated based on the tailored pulse sequence, and generating, by the machine learning model, quantitative histological maps based on the MR signals. In Some embodiments, the method 300 includes receiving qMR parameters based on MR signals generated based on the tailored pulse sequence, and generating, by the machine learning model, quantitative histological maps based on the qMR parameters. In some embodiments, the method 300 includes generating MR images based on the tailored pulse sequence, wherein the MR images have an image contrast associated with the biological parameters.Experimental Methods

[0038] The following section describes experimental methods relating to new qMR techniques and virtual histology for the non-invasive characterization of articular cartilage. These examples are merely illustrations and should not be construed as limiting the disclosure.

[0039] FIGS. 4-9 illustrate examples and results relating to qMR techniques and virtual histology for non-invasive characterization of articular cartilage.Preliminary Work

[0040] Validation of qMR’s sensitivity to histological parameters. Findings relating to the sensitivity of qMR parameters to cartilage structure and constituents were validated with quantitative histology. Existing MRI data from previous studies is used, which include various combinations of qMR maps, such as Ti, T2, Tirho, adiabatic Tirho, TRAFF, as well as quantitative histological reference data (FIG. 4). All cartilage specimens are processed using digital densitometry histology (proteoglycans) and polarized light microscopy (collagen orientation and anisotropy), as well as standard histology staining. The data shows a good correspondence between the qMR and the histology parameters.

[0041] Machine learning for the prediction of cartilage composition and structure. Prediction of histological parameters (e.g., collagen orientation, proteoglycan content) from qMR parameters using machine learning was studied. For example, a recent study showed that machine learning classification could successfully use qMR data to predict the depth-wise proteoglycan content and depth-wise collagen fiber angle in degenerated and intact cartilage (FIG. 5). Preliminary attempts were made to reconstruct synthetic quantitative histology maps of cartilage from 3 T in vivo qMR data via machine learning (FIG. 6).

[0042] Rapid multiparametric mapping of articular cartilage. A 2D pulse sequence (“IRFF”) based on the principles of MRF that concatenates an inversion recovery (“IR”) pulse with two fast imaging steady-state precession (“FISP”) segments and two fast low-angle shot (“FLASH”) segments to simultaneously map Bi+, proton density, Ti and T2 in the hip cartilage at 3 T, was implemented in a clinically feasible scan time of 7 minutes (FIG. 7). The technique relied on radial k-space sampling to achieve motion robustness. The same pulse sequence was modified to encode also TLho, was implemented as a 3D version, and applied for cartilage qMR in both the knee and the hip (FIG. 7). The IRFF sequence serves as the starting point in developing a new pulse sequence to estimate the subset of qMR parameters that best correspond to histology.I. Method for Histological Characterization of Cartilage from Quantitative MR

[0043] Data from the IRFF and an associated deep-learning model is used to generate synthetic quantitative histology maps of proteoglycans and collagen.Developing a neural network based on existing cartilage specimen data and tailorins the IRFF MRF pulse sequence for virtual histology

[0044] Deep Learning Model. The multiparametric qMR and quantitative histological data from previous experiments may be combined into a machine learning-based statistical methodology to predict histological parameters from qMR parameters (FIGS. 6 and 7). Such predictive models can be built using MLP neural networks or multivariate regression methods. However, even more suitable methods are Gaussian process-based methods, which have the advantage that the (stochastic) biophysical model (from Bloch equations) can be directly incorporated into the machine learning model following the physics-informed machine learning paradigm. A Gaussian process (“GP”)-based regression model is used to predict the histological parameters using a manually engineered set of features.

[0045] A U-Net may be used which is modified to be suitable for regression problems. However, in practice, the design of the architecture of the deep neural network, as well as the possible preprocessing steps needed for it, will heavily depend on the learnings from the Gaussian process-based methods that are developed first. It is possible that for robust operation, we will need to integrate physics-informed concepts, e.g., MR signal equations associated with qMR parameters quantification, into the deep learning model. The physics information may be integrated by using a combination of simulations and real data so that the backbone model is trained with a physical model, and it is then fine-tuned using real data.

[0046] Model Architecture and Training Strategy. The Gaussian process model may be implemented in GPflow, which is a Python-based Gaussian process library built on top of TensorFlow. In the kernel selection, the kernel was formed as a linear combination of constant kernels, linear kernels, and one of the following: squared exponential kernel, Matern kernel with v=3 / 2, or Matern kernel with v=5 / 2. Simple sparse Gaussian process methods are used and then sparse variational methods and Hilbert space methods if the need arises. The kernel parameters are first learned using maximum likelihood method (using the BFGS optimizer) and then the optimization is replaced with a Markov chain Monte Carlo (“MCMC”) method to enable proper uncertainty quantification.

[0047] The machine learning methodology is based on a deep learning architecture where the backbone model (a modified U-Net ) is trained using a large set of simulated data obtained using the MRF Bloch simulation environment used to design the pulse sequences. The modified U-Net is implemented using TensorFlow’ s Keras library and trained using standard optimizers available in the library, starting with the Adam optimizer. The backbone is improved by adding perturbations as well as via different approaches of incorporating histological properties, such as by reversing the models realized with the Gaussian process-based approach and via physical modeling. This use of simulated data enables circumvention of the issue of deep learning methods typically needing a large quantity of data compared to classical approaches. The first layers of the simulated-data-trained deep neural network are frozen, and the network will be fine-tuned using real data. By comparison, an optimal subset of four qMR parameters is identified and the resulting deep learning model (“DLi”) used to predict histological properties from those qMR parameters.

[0048] IRFF MRF Pulse Sequence. IRFF MRF pulse sequence has been extended to estimate additional qMR parameters in cartilage, such as Tirho and the parameters that characterize MT. Other work has shown that it is possible to tailor MRF pulse sequences to quantify different parameters. An IRFF MRF pulse sequence is used to estimate the optimal subset of four qMR parameters identified by the DLi model. The IRFF sequence has a modular design in which different building blocks are concatenated to encode multiple properties.MR pulse sequences tailored for the estimation of cartilage constituents

[0049] Existing optimization frameworks for pulse sequence optimization can be extended to optimize MR data acquisition directly for the estimation of cartilage constituents. An example for an existing framework uses the Cramer-Rao bound (“CRB) of the qMR parameters as”) a proxy for the noise variance in the estimated parameter. The CRB of one or multiple qMR parameters isused as the optimization objective that is minimized by a standard optimization algorithm. In a potential extension, the neural network that maps from the qMR parameters to the histology parameters is incorporated into the sequences design framework and the CRB of proteoglycans and / or collagen, rather than qMR parameters, is used as the optimization objective. One way to facilitate such an optimization, the neural network is extended to map the biological components back to the qMR parameters, as depicted in FIG. 8. To ensure bijectivity, additional free parameters fi are introduced that complement the biological ones but have no known biological meaning. The number of free parameters is chosen such that the number of qMR parameters matches the number of biological ones plus the number of free parameters. The overall network architecture loosely resembles that of autoencoder networks.

[0050] Using the acquired qMR data and the encoder part of the neural network (left half of FIG. 9), the distribution of the biological (proteoglycans and collagen) and free parameters found in the scanned cartilage samples is calculated. Using the decoder part of the network (right half of FIG. 9), for samples of the calculated distribution, the associated qMR parameters and, ultimately, the MR signal are simulated. The Cramer-Rao lower bound of the histology parameters is calculated and the pulse sequence is optimized by minimizing it by taking the derivative of the MR signal with respect to the qMR parameters using an established optimal control framework and then, following the chain rule, with respect to the histology parameters. The derivative with respect to the biological parameters is calculated with automatic differentiation of the neural network in resemblance of “physics-informed neural network” approaches. Finally, a reconstruction pipeline that concatenates the neural network-based model fitting with DLi and DL2 to generate synthetic histology maps from the MR signals is assembled.Definitions

[0051] The term “or,” as used herein, is used in its inclusive sense (and not in its exclusive sense) so that when used to connect a list of elements, the term “or” means one, some, or all of the elements in the list. Conjunctive language such as the phrase “at least one of X, Y, and Z,” unless specifically stated otherwise, is understood to convey that an element may be either X, Y, Z; X and Y; X and Z; Y and Z; or X, Y, and Z (i.e., any combination of X, Y, and Z). Thus, such conjunctive language is not generally intended to imply that certain embodiments require at least one of X, at least one of Y, and at least one of Z to each be present, unless otherwise indicated.

[0052] The hardware and data processing components used to implement the various processes, operations, illustrative logics, logical blocks, modules and circuits described inconnection with the embodiments disclosed herein may be implemented or performed with a general purpose single- or multi-chip processor, a graphical processing unit (GPU), a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA), or other programmable logic device, discrete gate or transistor logic, discrete hardware components, or any combination thereof designed to perform the functions described herein. A general purpose processor may be a microprocessor, or, any conventional processor, controller, microcontroller, or state machine. A processor also may be implemented as a combination of computing devices, such as a combination of a DSP and a microprocessor, a plurality of microprocessors, one or more microprocessors in conjunction with a DSP core, or any other such configuration. In some embodiments, particular processes and methods may be performed by circuitry that is specific to a given function. The memory (e.g., memory, memory unit, storage device) may include one or more devices (e.g., RAM, ROM, Flash memory, hard disk storage) for storing data and / or computer code for completing or facilitating the various processes, layers and modules described in the present disclosure. The memory may be or include volatile memory or non-volatile memory, and may include database components, object code components, script components, or any other type of information structure for supporting the various activities and information structures described in the present disclosure. According to an exemplary embodiment, the memory is communicably connected to the processor via a processing circuit and includes computer code for executing (e.g., by the processing circuit or the processor) the one or more processes described herein.

[0053] The present disclosure contemplates systems on any machine-readable media for accomplishing various operations. The embodiments of the present disclosure may be implemented using existing computer processors, or by a special purpose computer processor for an appropriate system, incorporated for this or another purpose, or by a hardwired system. Embodiments within the scope of the present disclosure include program products comprising machine-readable media for carrying or having machine-executable instructions or data structures stored thereon. Such machine-readable media can be any available media that can be accessed by a general purpose or special purpose computer or other machine with a processor. By way of example, such machine-readable media can comprise RAM, ROM, erasable programmable readonly memory (EPROM), electronically erasable programmable read-only memory (EEPROM), or other optical disk storage, magnetic disk storage or other magnetic storage devices, or any other medium which can be used to carry or store desired program code in the form of machineexecutable instructions or data structures and which can be accessed by a general purpose or special purpose computer or other machine with a processor. Combinations of the above are alsoincluded within the scope of machine-readable media. Machine-executable instructions include, for example, instructions and data which cause a general-purpose computer, special-purpose computer, or special-purpose processing machines to perform a certain function or group of functions.

[0054] It is important to note that the construction and arrangement of the system as shown in the various exemplary embodiments is illustrative only. Additionally, any element disclosed in one embodiment may be incorporated or utilized with any other embodiment disclosed herein.

Claims

WHAT IS CLAIMED IS:

1. A system for generating quantitative maps of biological parameters, comprising: a processor; and a memory coupled to the processor, the memory having instructions stored thereon which, when executed by the processor, cause the system to: generate, using one or more machine learning models, a tailored pulse sequence based on one or more biological parameters associated with a tissue of interest; send the tailored pulse sequence to a magnetic resonance (MR) scanner; receive MR signals for the tissue of interest, the MR signals associated with the tailored pulse sequence; and generate, using the one or more machine learning models, quantitative histology maps from the MR signals.

2. The system of claim 1, wherein the instructions further cause the system to train the one or more machine learning models based on MRI data relating to the tissue of interest, and quantitative magnetic resonance (qMR) maps and quantitative histological data associated with the MRI data.

3. The system of claim 2, wherein the one or more machine learning models is trained to map from qMR parameters to histology parameters.

4. The system of claim 1, wherein the instructions further cause the system to train the machine learning models based on the generated quantitative histology maps.

5. The system of claim 1, wherein the instructions further cause the system to identify, by the one or more machine learning models, a combination of qMR parameters associated with the one or more biological parameters associated with the tissue of interest.

6. The system of claim 1, wherein the one or more machine learning models includes a neural network.

7. The system of claim 1, wherein the instructions further cause the system to receive qMR parameters associated with the MR signals.

8. A non-transitory processor-readable medium storing code representing instructions to be executed by one or more processors, the instructions comprising code to cause the one or more processors to: generate, using one or more machine learning models, a tailored pulse sequence sensitive to one or more biological parameters associated with a tissue of interest, the one or more machine learning models trained to optimize the tailored pulse sequence based on an optimization objective; send the tailored pulse sequence to an MR scanner; receive MR signals for the tissue of interest, the MR signals associated with the tailored pulse sequence; and generate, using the one or more machine learning models, quantitative histology maps from the MR signals.

9. The non-transitory processor-readable medium of claim 8, wherein the instructions further comprise code to cause the one or more processors to train the one or more machine learning models based on MRI data relating to the tissue of interest, and quantitative magnetic resonance (qMR) maps and quantitative histological data associated with the MRI data.

10. The non-transitory processor-readable medium of claim 8, wherein the one or more machine learning models includes a neural network.

11. The non-transitory processor-readable medium of claim 8, wherein the instructions further comprise code to cause the one or more processors to train the one or more machine learning models based on the generated quantitative histology maps.

12. The non-transitory processor-readable medium of claim 8, wherein the one or more machine learning models is trained to map from qMR parameters to histology parameters.

13. A method of optimizing a magnetic resonance (MR) pulse sequence for generating quantitative maps of biological parameters, the method comprising: receiving, by a processor, quantitative magnetic resonance (qMR) parameters relating to a tissue of interest; and generating, by one or more machine learning models based on the qMR parameters and an optimization objective, a tailored pulse sequence sensitive to biological parameters associated with the tissue of interest.

14. The method of claim 13, wherein the optimization objective includes increased sensitivity to the biological parameters.

15. The method of claim 13, further comprising: receiving, by the processor, MR signals generated responsive to the tailored pulse sequence; and generating, by the one or more machine learning models, quantitative histological maps based on the MR signals.

16. The method of claim 13, wherein the optimization objective is a Cramer-Rao bound (CRB) of one or more of the qMR parameters.

17. The method of claim 13, wherein one or more machine learning models includes a neural network.

18. The method of claim 13, further comprising generating MR images based on the tailored pulse sequence, the MR images having an image contrast associated with the biological parameters.

19. The method of claim 13, wherein the qMR parameters are determined based on magnetic resonance fingerprinting.

20. The method of claim 13, further comprising: receiving, by the processor, qMR parameters based on MR signals generated; and generating, by the one or more machine learning models, quantitative histological maps based on the qMR parameters.

Images