Fully automated detection of focal cortical dysplasia: Comparison of MPRAGE and MP2RAGE sequences

Abstract

Objective: The detection of focal cortical dysplasia (FCD) in magnetic resonance imaging is challenging. Voxel-based morphometric analysis and automated FCD detection using an artificial neural network (ANN) integrated into the Morphometric Analysis Program (MAP18) have been shown to facilitate FCD detection. This study aimed to evaluate whether the detection of FCD can be further improved by feeding this approach with magnetization prepared two rapid acquisition gradient echoes (MP2RAGE) instead of magnetization-prepared rapid acquisition gradient echo (MPRAGE) datasets.Methods: MPRAGE and MP2RAGE datasets were acquired in a consecutive sample of 32 patients with FCD and postprocessed using MAP18. Visual analysis and, if available, histopathology served as the gold standard for assessing the sensitivity and specificity of FCD detection. Out-of-sample specificity was evaluated in a cohort of 32 healthy controls. Results:The sensitivity and specificity of FCD detection were 82.4% and 62.5% for the MPRAGE and 97.1% and 34.4% for the MP2RAGE sequences, respectively.Median volumes of true-positive voxel clusters were .16 ml for the MPRAGE and .52 ml for the MP2RAGE sequences compared to .08-and .04-ml volumes of falsepositive clusters. With regard to cluster volumes, FCD detection was substantially improved for the MP2RAGE data when the estimated optimal threshold of .23 ml was applied (sensitivity = 72.9%, specificity = 83.0%). In contrast, the estimated optimal threshold of .37 ml for the MPRAGE data did not improve FCD lesion detection (sensitivity = 42.9%, specificity = 79.5%).Significance: In this study, the sensitivity of FCD detection by morphometric analysis and an ANN integrated into MAP18 was higher for MP2RAGE than for MPRAGE sequences. Additional usage of cluster volume information helped to discriminate between true-and false-positive MP2RAGE results.

Artificial intelligence for the detection of focal cortical dysplasia: Challenges in translating algorithms into clinical practice

“…Most approaches formulate the problem of FCD detection as a classification task. 13,16,24,34,35,40,46,69,70,[71][72][73][74] Input data range from raw MRI data to morphometric maps or surface features. They can be one-dimensional, that is, single voxels (or vertices if the input data are surface-based), or two-or three-dimensional images.…”

“…Many works do not uniquely belong to one of these three categories and involve a mix of other processing steps. For example, several approaches use morphometric maps and other differences compared to a “normal” cohort as inputs to a classification model 13,24,71,73 . Colliot et al 83 explore segmentation with coarse localization information as additional input, which could be helpful for hypothesis refinement.…”

“…In the two-and three-dimensional cases, an algorithm often only operates on smaller image parts, socalled "patches." The output is thus generated on the voxel level, 24,46,[71][72][73] vertex level, 13,18,35,40,70 or patch level. 16,34,69,74 Figure 2 shows examples of the outputs from Multi-centre Epilepsy Lesion Detection (MELD), 18 Morphometric Analysis Program v2018 (MAP18), 24 and deepFCD 16 models.…”

“…Its output is a value between 0 and 1, akin to the probability for the input data belonging to a particular class, as shown in Figure 3A,B. For FCD detection, the possible classes are usually either “lesional” or “nonlesional.” Most approaches formulate the problem of FCD detection as a classification task 13,16,24,34,35,40,46,69,70,71–74 . Input data range from raw MRI data to morphometric maps or surface features.…”

Pushing MP2RAGE boundaries: Ultimate time‐efficient parameterization combined with exhaustive T1 synthetic contrasts

“…These parameters were initially optimized by Marques et al to obtain the highest and most uniform T 1 ‐weighted brain tissue contrast when combining the two acquired images in a complex ratio. This so‐called “UNI” contrast, free from $$$ {\mathrm{B}}_1^{-} $$$ reception bias, showed high potential for segmentation purposes 1,3 and major interest in pathological contexts such as epilepsy or multiple sclerosis (MS) imaging at 7 T 4–8 and 3 T, 9,10 progressively turning into a surrogate to the MPRAGE sequence for T 1 ‐weighted brain imaging 2 . Moreover, from the UNI image, a quantitative T 1 map 11 can be derived that can be useful in clinical research 12–14 …”