Creative Commons source reading · Image guidance and surgical robotics
Deep learning-based Intraoperative MRI reconstruction.
Jon André Ottesen; Tryggve Storas; Svein Are Sirirud Vatnehol; Grethe Løvland; Einar Osland Vik-Mo; Till Schellhorn; Karoline Skogen; Christopher Larsson; Atle Bjørnerud; Inge Rasmus Groote-Eindbaas; Matthan W A Caan
Originally published in European radiology experimental · 2025-02-25
From the article
Selected source text. JournalOS removes markup and shortens excerpts where indicated. The full article provides the complete context.
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.
Results
Averaged across all metrics, the DL reconstructions were rated with a higher score than CS reconstruction for 72%, 72%, and 14% of the ratings, for readers 1, 2, and 3, respectively. Conversely, the CS reconstruction was preferred over DL reconstruction for 8%, 7%, and 37% of the ratings, respectively for readers 1, 2, and 3. The violin plot distribution of the qualitative assessments is shown in Fig. 2. Of note, since the assessment criteria were discrete, no interpolation between the metrics was performed.Fig.
Discussion
In this study, we have adapted and trained a DL reconstruction network to reconstruct on-scanner undersampled iMRI data. The model was qualitatively evaluated and compared to the CS reconstructions by three expert readers on a cohort of 40 patients who underwent iMRI. DL was found to reconstruct iMRI with significantly better image quality compared to CS for two of the three readers. Notably, the neuroradiologists (readers 1 and 2) favored overall DL over CS to a larger extent than the neurosurgeon (reader 3). Reader 3 assessed that the CS reconstruction had improved image quality compared to the DL equivalent.
Editorial reading guide
- How were patients or cases selected, and what did the authors use as a comparator?
- Which outcomes, complications and follow-up intervals determine how far the findings apply?
