Estimation of the capillary level input function for dynamic contrast‐enhanced MRI of the breast using a deep learning approach

Abstract

Purpose: To develop a deep learning approach to estimate the local capillarylevel input function (CIF) for pharmacokinetic model analysis of DCE-MRI. Methods:A deep convolutional network was trained with numerically simulated data to estimate the CIF. The trained network was tested using simulated lesion data and used to estimate voxel-wise CIF for pharmacokinetic model analysis of breast DCE-MRI data using an abbreviated protocol from women with malignant (n = 25) and benign (n = 28) lesions. The estimated parameters were used to build a logistic regression model to detect the malignancy. Result:The pharmacokinetic parameters estimated using the network-predicted CIF from our breast DCE data showed significant differences between the malignant and benign groups for all parameters. Testing the diagnostic performance with the estimated parameters, the conventional approach with arterial input function (AIF) showed an area under the curve (AUC) between 0.76 and 0.87, and the proposed approach with CIF demonstrated similar performance with an AUC between 0.79 and 0.81. Conclusion:This study shows the feasibility of estimating voxel-wise CIF using a deep neural network. The proposed approach could eliminate the need to measure AIF manually without compromising the diagnostic performance to detect the malignancy in the clinical setting.

Feasibility of measuring blood‐brain barrier permeability using ultra‐short echo time radial magnetic resonance imaging

“…31 As an alternative, to improve the uncertainty coming from the AIF selection, we seek to implement deep-learning based DCE analysis, which has recently shown to estimate BBB permeability with reduced scan time. 32,33 Also, our radial imaging sequence allows retrospective selection of temporal resolution, and previous studies have already demonstrated this flexibility for different applications. 9,34 While it is not presented here, we investigated the effect of reconstructing dynamic images at different temporal resolutions and identified the optimal temporal resolution for appropriately selecting AIF and capturing contrast agent dynamics.…”

“…The accurate AIF estimation is essential for accurate parameter estimation and it needs to be carefully examined between 3D UTE‐GRASP method and the conventional sequence, as different undersampling in Cartesian acquisition has shown to degrade the temporal fidelity 31 . As an alternative, to improve the uncertainty coming from the AIF selection, we seek to implement deep‐learning based DCE analysis, which has recently shown to estimate BBB permeability with reduced scan time 32,33 . Also, our radial imaging sequence allows retrospective selection of temporal resolution, and previous studies have already demonstrated this flexibility for different applications 9,34 .…”

Improving measurement of blood-brain barrier permeability with reduced scan time using deep-learning-derived capillary input function

“…However, direct measurement of the capillary-level GBCA concentration changes are not feasible due to the limited spatial resolution in MRI. Thus, we propose to utilize a deep learning network to predict the CIF that is local to a given patch of the tissue contrast dynamics, as demonstrated the feasibility with breast DCE-MRI data recently ( Bae et al, 2022 ). To model the vascular transport from the upstream artery to the capillary, we adopted two different pharmacokinetic models with different vascular transport designs.…”

“…We also conducted another study to overcome the challenges associated with the uncertainty in the AIF selection. We utilized the power of deep learning networks to automatically estimate the local capillary input function at the voxel-level for pharmacokinetic model analyses from the tissue contrast enhancement data ( Bae et al, 2022 ). In this study, we trained a network to predict the local capillary-level input function (CIF) for a given patch of the tissue contrast dynamics.…”

“…Adopting these developments from the previous two studies ( Bae and al., 2018 , Bae et al, 2022 ), this study was conducted to improve quantification of BBB permeability in normal aging while reducing the scan time to 10 min. Using the trained deep learning network for CIF, we aim to show that the predicted CIFs provide accurate BBB permeability measurements comparable to the reference of using the full 25 min scan with the AIF.…”