Deep learning-based automated segmentation of resection cavities on postsurgical epilepsy MRI

Arnold, Thomas C.; Muthukrishnan, Ramya; Pattnaik, Akash R; Sinha, Nishant; Gibson, Adam; Gonzalez, Hannah; Das, Sandhitsu R.; Litt, Brian; Englot, Dario J.; Morgan, Victoria L.; Davis, Kathryn A.; Stein, Joel M.

doi:10.1016/j.nicl.2022.103154

Cited by 8 publications

(17 citation statements)

References 36 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…As previously reported in other studies (13,22), we found a relationship between DSC and lacuna size indicating that there is a trend towards better performance for larger resection. Since DSC is dependent upon a volume ratio, large values in the numerator (intersection) and denominator (total volume) will be less sensitive to intersection mismatches when comparing two volumetric shapes.…”

Section: Discussionsupporting

confidence: 90%

“…Among the fully automatic tools we tested, DeepResection was the fastest (~35 s per image), ResectVol DL was second (~42 s), whereas ResectVol 1.1.2 took much longer (~9 min) due to the nature of its processing algorithm which is based on a series of processing steps. As previously reported in other studies (13,22), we found a relationship between DSC and lacuna size indicating that there is a trend towards better performance for larger resection. Since DSC is dependent upon a volume ratio, large values in the numerator (intersection) and denominator (total volume) will be less sensitive to intersection mismatches when comparing two volumetric shapes.…”

Section: Discussionsupporting

confidence: 89%

“…The model training is carried out with five cross-validation folds and 1000 epochs for each fold. To compare our algorithm against other freely available tools, we also performed the lacuna prediction with two other approaches: ResectVol version 1.1.2 (freely available at lniunicamp.com/resectvol), a rule-based algorithm developed by our group (Casseb et al, 2021); and the DeepResection tool (available at github.com/penn-cnt/DeepResection) developed by Arnold and colleagues (Arnold et al, 2022) that uses a set of three 2D U-Nets, each in one anatomical plane, followed by voxel-wise majority voting. In ResectVol 1.1.2, it is mandatory for users to provide the pre-and postoperative images to run the analysis, whereas in DeepResection, users are required to provide the postoperative image only.…”

Section: Automatic Segmentationmentioning

confidence: 99%

“…Characterization of the lacuna can be informative to those predictive models by providing volume and location of extracted brain tissue. Due to the laborious and time-consuming nature of the manual annotation required to characterize resections, initiatives to automate this task increased in the last years (11)(12)(13). Hence, the purpose of this study is twofold: (i) to introduce ResectVol DL, a fully automatic method to segment brain lacunas, based on the 3D U-Net architecture (14); and (ii) to compare ResectVol DL against two other methods: ResectVol (version 1.1.2), a rule-based processing pipeline (12), and DeepResection, which is based on a 2D U-Net architecture (13).…”

Section: Introductionmentioning

confidence: 99%

“…Since manual segmentation is tedious and time-consuming, initiatives to automate this task increased in the last years (Arnold et al, 2022; Billardello et al, 2022; Casseb et al, 2021). In this study, we describe ResectVol DL, which is based on the 3D U-Net architecture using the nnU-Net framework (Isensee et al, 2020), tested on resective lacunas, and compared against two other methods: ResectVol (version 1.1.2), a rule-based method previously published by our team (Casseb et al, 2021), and DeepResection, a 2D U-Net implementation by Arnold and colleagues (Arnold et al, 2022). Our goal is to develop an open-source platform to enable surgical epilepsy research with optimal precision of the surgery definition.…”

Section: Introductionmentioning

confidence: 99%

See 4 more Smart Citations

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

Casseb,

de Campos,

Loos

et al. 2023

Preprint

View full text Add to dashboard Cite

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.

show abstract

Section: Discussionsupporting

confidence: 90%

Section: Discussionsupporting

confidence: 89%