Deep learning‐based reconstruction of in vivo pelvis conductivity with a 3D patch‐based convolutional neural network trained on simulated MR data

Abstract

Purpose To demonstrate that mapping pelvis conductivity at 3T with deep learning (DL) is feasible. Methods 210 dielectric pelvic models were generated based on CT scans of 42 cervical cancer patients. For all dielectric models, electromagnetic and MR simulations with realistic accuracy and precision were performed to obtain and transceive phase ( ϕ ± ). Simulated and ϕ ± served as input to a 3D patch‐based convolutional neural network, which was trained in a supervised fashion to retrieve the conductivity. The same network architecture was retrained using only ϕ ± in input. Both network configurations were tested on simulated MR data and their conductivity reconstruction accuracy and precision were assessed. Furthermore, both network configurations were used to reconstruct conductivity maps from a healthy volunteer and two cervical cancer patients. DL‐based conductivity was compared in vivo and in silico to Helmholtz‐based (H‐EPT) conductivity. Results Conductivity maps obtained from both network configurations were comparable. Accuracy was assessed by mean error (ME) with respect to ground truth conductivity. On average, ME < 0.1 Sm −1 for all tissues. Maximum MEs were 0.2 Sm −1 for muscle and tumour, and 0.4 Sm −1 for bladder. Precision was indicated with the difference between 90 th and 10 th conductivity percentiles, and was below 0.1 Sm −1 for fat, bone and muscle, 0.2 Sm −1 for tumour and 0.3 Sm −1 for bladder. In vivo, DL‐based conductivity had median values in agreement with H‐EPT values, but a higher precision. Conclusion Anatomically detailed, noise‐robust 3D conductivity maps with good sensitivity to tissue conductivity variations were reconstructed in the pelvis with DL.

Radio-Frequency Vector Magnetic Field Mapping in Magnetic Resonance Imaging

“…Such a problem might be satisfactorily handled with various regularization or machine learning techniques [27] to supplement available equations. For the latter, RF simulation with detailed human body model [28] could generate training data that link limited, measurable B1 components with full RF vector solution. Instead of reducing angles in a conventional cylindrical magnet, it is also conceivable to implement large B0-B1 reorientation by rotating B0 with a portable magnet.…”

Hyperthermia Treatment Planning: Clinical Application and Ongoing Developments

“…Recently, efficient deep-learning based methods were investigated, in which a convolutional neural network (CNN) is trained to learn the mapping relation between B1 + and electric properties [81,82]. In silico validation and in vivo comparison with a conventional EPT technique demonstrated that deep-learning EPT in the pelvis yields anatomically-detailed and noise-robust 3D conductivity maps with good sensitivity to tissue conductivity variations [82].…”

Technical developments for MR-based electrical property mapping

“…Including dl-ept to obtain the patient-specific conductivity represents an important progress towards reliable hyperthermia treatment planning. We have recently shown that dl-based pelvis conductivity maps display high reconstruction accuracy at tissue interfaces, robustness against noise in measured B + 1 maps and acceptable sensitivity to conductivity variations [241]. In contrast to dl-ept, the present state-of-art method for in vivo pelvis conductivity mapping [57,102,131], i.e.…”

“…Recently, Gavazzi et al [241] have developed a dl-ept approach for pelvis conductivity mapping where a convolutional neural network infers the conductivity of pelvic tissues from |B + 1 | and φ ± maps measured with mri, after being trained on in silico dielectric pelvic models with realistic conductivity values at 128 MHz. For implementation details, validation and comparison with a conventional ept method, the reader is referred to Chapter 4.…”

“…In dl-ept, electrical properties of a body site at a specific field strength are reconstructed by a convolutional neural network trained on large datasets of simulated B + 1 distributions. Very recently, a novel dl-ept method has been developed for reconstruction of in vivo 3D conductivity maps in the pelvic region [241]. Both the in silico validation and the in vivo comparison with a conventional ept technique demonstrated that this dl-ept method is able to reconstruct pelvis conductivity maps with high noise-robustness and accuracy at tissue interfaces.…”