Convolutional Neural Networks for Direct Inference of Pharmacokinetic Parameters: Application to Stroke Dynamic Contrast-Enhanced MRI

Abstract

Background and Purpose: The T1-weighted dynamic contrast enhanced (DCE)-MRI is an imaging technique that provides a quantitative measure of pharmacokinetic (PK) parameters characterizing microvasculature of tissues. For the present study, we propose a new machine learning (ML) based approach to directly estimate the PK parameters from the acquired DCE-MRI image-time series that is both more robust and faster than conventional model fitting.Materials and Methods: We specifically utilize deep convolutional neural networks (CNNs) to learn the mapping between the image-time series and corresponding PK parameters. DCE-MRI datasets acquired from 15 patients with clinically evident mild ischaemic stroke were used in the experiments. Training and testing were carried out based on leave-one-patient-out cross- validation. The parameter estimates obtained by the proposed CNN model were compared against the two tracer kinetic models: (1) Patlak model, (2) Extended Tofts model, where the estimation of model parameters is done via voxelwise linear and nonlinear least squares fitting respectively.Results: The trained CNN model is able to yield PK parameters which can better discriminate different brain tissues, including stroke regions. The results also demonstrate that the model generalizes well to new cases even if a subject specific arterial input function (AIF) is not available for the new data.Conclusion: A ML-based model can be used for direct inference of the PK parameters from DCE image series. This method may allow fast and robust parameter inference in population DCE studies. Parameter inference on a 3D volume-time series takes only a few seconds on a GPU machine, which is significantly faster compared to conventional non-linear least squares fitting.

Comparison of iterative parametric and indirect deep learning‐based reconstruction methods in highly undersampled DCE‐MR Imaging of the breast

“…While iterative methods exist for directly estimating TK parameters from undersampled data, 19 deep learning methods for directly estimating TK parameters have also been attempted before. 23,24 These approaches though perform at par with model-based iterative parameter estimations, have major limitations because of the fully connected layers and stacking the time dimension data as channels in input. Due to these two factors, the network cannot be deployed to scans with different number of time points and/or image dimensions.…”

“…However, instead of image as input, they used concentration maps and AIF as input. Cagdas et al 23,24 have estimated the tracer-kinetic parameters through DL using dilated convolution and fully connected layers, where the time dimension was treated as channels. Limitation of such method is the lack of robustness for data with different number of time volumes and inflexibility in choice of PK models.…”

“…However, instead of image as input, they used concentration maps and AIF as input. Cagdas et al 23,24 . have estimated the tracer‐kinetic parameters through DL using dilated convolution and fully connected layers, where the time dimension was treated as channels.…”

Exercise‐induced calf muscle hyperemia: Rapid mapping of magnetic resonance imaging using deep learning approach

“…A NN has the additional benefit of mitigating convergence issues of model fitting as well as nearly instantaneous inference. The promise of the method has been demonstrated in recent study with patients of mild ischemic stroke (Ulas, Das, & Thrippleton, 2018). DCE MRI of skeletal muscles often consists of multiple scans stimulated by exercise of different intensities (Zhang et al., 2019).…”

Efficient DCE-MRI Parameter and Uncertainty Estimation Using a Neural Network

“…Such methods typically use optimization algorithms and multiple initializations to solve issues arising from regions of complete or approximate flatness in the cost function landscape [6], [9], [10]. More recently, neural networks have been proposed to shift the computational load from (frequent) inference to (one-time) training [11]. For these algorithms, measures of precision are obtained through either time-consuming Monte-Carlo simulations (MC) with multiple noise realizations and initializations, or variance estimation through linear error propagation [12].…”