Fully automatic segmentation of brain lacunas resulting from resective surgery using a 3D deep learning model.

The rapid and constant development of deep learning (DL) strategies is pushing forward the quality of object segmentation in images from diverse fields of interest. In particular, these algorithms can be very helpful in delineating brain abnormalities (lesions, tumors, lacunas, etc), enabling the extraction of information such as volume and location, that can inform doctors or feed predictive models. In this study, we describe ResectVol DL, a fully automatic tool developed to segment resective lacunas in brain images of patients with epilepsy. ResectVol DL relies on the nnU-Net framework that leverages the 3D U-Net deep learning architecture. T1-weighted MRI datasets from 120 patients (57 women; 31.5 ± 15.9 years old at surgery) were used to train (n=78) and test (n=48) our tool. Manual segmentations were carried out by five different raters and were considered as ground truth for performance assessment. We compared ResectVol DL with two other fully automatic methods: ResectVol 1.1.2 and DeepResection, using the Dice similarity coefficient (DSC), Pearson's correlation coefficient, and relative difference to manual segmentation. ResectVol DL presented the highest median DSC (0.92 vs. 0.78 and 0.90), the highest correlation coefficient (0.99 vs. 0.63 and 0.94) and the lowest median relative difference (9 vs. 44 and 12 %). Overall, we demonstrate that ResectVol DL accurately segments brain lacunas, which has the potential to assist in the development of predictive models for postoperative cognitive and seizure outcomes.

Multi-Coil MRI Reconstruction Challenge-Assessing Brain MRI Reconstruction Models and Their Generalizability to Varying Coil Configurations.

Deep-learning-based brain magnetic resonance imaging (MRI) reconstruction methods have the potential to accelerate the MRI acquisition process. Nevertheless, the scientific community lacks appropriate benchmarks to assess the MRI reconstruction quality of high-resolution brain images, and evaluate how these proposed algorithms will behave in the presence of small, but expected data distribution shifts. The multi-coil MRI (MC-MRI) reconstruction challenge provides a benchmark that aims at addressing these issues, using a large dataset of high-resolution, three-dimensional, T1-weighted MRI scans. The challenge has two primary goals: (1) to compare different MRI reconstruction models on this dataset and (2) to assess the generalizability of these models to data acquired with a different number of receiver coils. In this paper, we describe the challenge experimental design and summarize the results of a set of baseline and state-of-the-art brain MRI reconstruction models. We provide relevant comparative information on the current MRI reconstruction state-of-the-art and highlight the challenges of obtaining generalizable models that are required prior to broader clinical adoption. The MC-MRI benchmark data, evaluation code, and current challenge leaderboard are publicly available. They provide an objective performance assessment for future developments in the field of brain MRI reconstruction.

Extraction of a vascular function for a fully automated dynamic contrast-enhanced magnetic resonance brain image processing pipeline.

PURPOSE: To develop a deep-learning model that leverages the spatial and temporal information from dynamic contrast-enhanced magnetic resonance (DCE MR) brain imaging in order to automatically estimate a vascular function (VF) for quantitative pharmacokinetic (PK) modeling. METHODS: Patients with glioblastoma multiforme were scanned post-resection approximately every 2 months using a high spatial and temporal resolution DCE MR imaging sequence ( ≈5 s and ≈2 cm3 ). A region over the transverse sinus was manually drawn in the dynamic T1-weighted images to provide a ground truth VF. The manual regions and their resulting VF curves were used to train a deep-learning model based on a 3D U-net architecture. The model concurrently utilized the spatial and temporal information in DCE MR images to predict the VF. In order to analyze the contribution of the spatial and temporal terms, different weighted combinations were examined. The manual and deep-learning predicted regions and VF curves were compared. RESULTS: Forty-three patients were enrolled in this study and 155 DCE MR scans were processed. The 3D U-net was trained using a loss function that combined the spatial and temporal information with different weightings. The best VF curves were obtained when both spatial and temporal information were considered. The predicted VF curve was similar to the manual ground truth VF curves. CONCLUSION: The use of spatial and temporal information improved VF curve prediction relative to when only the spatial information is used. The method generalized well for unseen data and can be used to automatically estimate a VF curve suitable for quantitative PK modeling. This method allows for a more efficient clinical pipeline and may improve automation of permeability mapping.

Dual-domain cascade of U-nets for multi-channel magnetic resonance image reconstruction.

The U-net is a deep-learning network model that has been used to solve a number of inverse problems. In this work, the concatenation of two-element U-nets, termed the W-net, operating in k-space (K) and image (I) domains, were evaluated for multi-channel magnetic resonance (MR) image reconstruction. The two-element network combinations were evaluated for the four possible image-k-space domain configurations: a) W-net II, b) W-net KK, c) W-net IK, and d) W-net KI. Selected four element (WW-nets) and six element (WWW-nets) networks were also examined. Two configurations of each network were compared: 1) each coil channel was processed independently, and 2) all channels were processed simultaneously. One hundred and eleven volumetric, T1-weighted, 12-channel coil k-space datasets were used in the experiments. Normalized root mean squared error, peak signal-to-noise ratio and visual information fidelity were used to assess the reconstructed images against the fully sampled reference images. Our results indicated that networks that operate solely in the image domain were better when independently processing individual channels of multi-channel data. Dual-domain methods were better when simultaneously reconstructing all channels of multi-channel data. In addition, the best cascade of U-nets performed better (p < 0.01) than the previously published, state-of-the-art Deep Cascade and Hybrid Cascade models in three out of four experiments.

Interactive patient-customized curvilinear reformatting for improving neurosurgical planning.

PURPOSE: Visualizing a brain in its native space plays an essential role during neurosurgical planning because it allows the superficial cerebral veins and surrounding regions to be preserved. This paper describes the use of a visualization tool in which single gadolinium contrast-enhanced T1-weighted magnetic resonance imaging was applied in nondefective and nonresective skulls to promote visualization of important structures. METHODS: A curvilinear reformatting tool was applied on the supratentorial compartment to peel the tissues to the depth of the dura mater and thereby revealing cortical and vascular spatial relationships. The major advantage of our proposed tool is that it does not require coregistration of anatomical and vascular volumes. RESULTS: The reliability of this technique was supported by comparisons between preoperative images and digital photographs of the brain cortical surface obtained after the dura mater was removed in 20 patients who underwent surgery in the Clinics Hospital of the University of Campinas from January 2017 to April 2018. CONCLUSION: Single fat-suppressed GAD contrast-enhanced T1-weighted magnetic resonance scans provide accurate preoperative 3D views of cortical and vascular relationships similar to neurosurgeons' intraoperative views. In developing countries with limited access to state-of-the-art health technologies, this imaging approach may improve the safety of complex neurosurgeries.

Toward a Multimodal Diagnostic Exploratory Visualization of Focal Cortical Dysplasia.

Focal cortical dysplasia (FCD) is a malformation of cortical development and a common cause of pharmacoresistant epilepsy. Resective surgery of clear-cut lesions may be curative. However, the localization of the seizure focus and the evaluation of its spatial extent can be challenging in many situations. For concordance assessment, medical studies show the relevance of accurate correlation of multisource imaging sequences. to improve the sensitivity and specificity of the evaluation. In this paper, we share the process we went through to reach our simple, but effective, solution for integrating multi-volume rendering into an exploratory visualization environment for the diagnosis of FCD. We focus on fetching of multiple data assigned to a sample when they are rendered. Knowing that the major diagnostic role of multiple volumes is to complement information, we demonstrate that appropriate geometric transformations in the texture space are sufficient for accomplishing this task. This allows us to fully implement our proposal in the OpenGL rendering pipeline and to easily integrate it into the existing visual diagnostic application. Both time performance and the visual quality of our proposal were evaluated with a set of clinical data volumes for assessing the potential practical impact of our solution in routine diagnostic use.