Deep learning-based Intraoperative MRI reconstruction

Abstract

Background: We retrospectively evaluated the quality of deep learning (DL) reconstructions of on-scanner accelerated intraoperative MRI (iMRI) during respective brain tumor surgery.

Methods: Accelerated iMRI was performed using dual surface coils positioned around the area of resection. A DL model was trained on the fastMRI neuro dataset to mimic the data from the iMRI protocol. The evaluation was performed on imaging material from 40 patients imaged from Nov 1, 2021, to June 1, 2023, who underwent iMRI during tumor resection surgery. A comparative analysis was conducted between the conventional compressed sense (CS) method and the trained DL reconstruction method. Blinded evaluation of multiple image quality metrics was performed by two neuroradiologists and one neurosurgeon using a 1-to-5 Likert scale (1, nondiagnostic; 2, poor; 3, acceptable; 4, good; and 5, excellent), and the favored reconstruction variant.

Results: The DL reconstruction was strongly favored or favored over the CS reconstruction for 33/40, 39/40, and 8/40 of cases for readers 1, 2, and 3, respectively. For the evaluation metrics, the DL reconstructions had a higher score than their respective CS counterparts for 72%, 72%, and 14% of the cases for readers 1, 2, and 3, respectively. Still, the DL reconstructions exhibited shortcomings such as a striping artifact and reduced signal.

Conclusion: DL shows promise in allowing for high-quality reconstructions of iMRI. The neuroradiologists noted an improvement in the perceived spatial resolution, signal-to-noise ratio, diagnostic confidence, diagnostic conspicuity, and spatial resolution compared to CS, while the neurosurgeon preferred the CS reconstructions across all metrics.

Relevance statement: DL shows promise to allow for high-quality reconstructions of iMRI, however, due to the challenging setting of iMRI, further optimization is needed.

Key points: iMRI is a surgical tool with a challenging image setting. DL allowed for high-quality reconstructions of iMRI. Additional optimization is needed due to the challenging intraoperative setting.

Keywords: Artifacts; Brain neoplasms; Deep learning; Magnetic resonance imaging; Neurosurgical procedures.

Fig. 1

Illustration of the raw iMRI data before any reconstruction, the model training/reconstruction regime, and the resulting DL reconstruction with and without bias field correction. The iMRI scans have two different imaging protocols a large FOV and a small FOV. A DL model was trained on the fastMRI dataset to match the iMRI protocol with respect to the masked k-space, FOVs, and resolution. The model was used to reconstruct the on-scanner iMRI scans followed by bias field correction

Fig. 2

A violin plot of the qualitative assessment of the DL and CS reconstructions for 40 intraoperative patients from three expert readers. The mean of the DL and CS assessments are shown with the colored lines. No smoothing has been used in the violin plot due to the discrete nature of the assessment

Fig. 3

The preferred reconstruction variant among the three expert readers. DL, Deep learning; CS, Compressed sense

Fig. 4

Representative examples of the DL and CS reconstructions from four different patients, depicting T1-weighted pre-contrast, T1-weighted post-contrast, and T2-weighted FLAIR scans. Window level was chosen between 0.05 and 0.995 percentiles

Fig. 5

Examples of DL-specific artifacts. This includes high noise levels for a given scan, striping-like artifacts, and high noise for one slice in an image

Fig. 6

Three cases highlighted by the neurosurgeon where the DL reconstructions had considerable artifacts and were evaluated as nondiagnostic or poor by the neurosurgeon compared to the compressed sensing (CS) counterpart. For the two last patients, the artifacts affected the area of resection for one or more of the scans. The first patient, i.e., columns 1 and 2 did not have a T1-weighed post-contrast series