Automatic Rodent Brain MRI Lesion Segmentation with Fully Convolutional Networks

Valverde, Juan Miguel; Shatillo, Artem; Feo, Riccardo De; Gröhn, Olli; Sierra, Alejandra; Tohka, Jussi

doi:10.1007/978-3-030-32692-0_23

Cited by 8 publications

(20 citation statements)

References 17 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Our fully-automatic approach is trained end to end, requires no preprocessing, and it was validated on a large and diverse data set composed by 916 MRI rat brain scans at nine different lesion stages from 671 rats utilized to study focal cerebral ischemia. We extend our earlier conference paper (Valverde et al, 2019 ) by (1) improving our previous ConvNet (Valverde et al, 2019 ) with a deeper and different architecture and providing an ablation study (Meyes et al, 2019 ) justifying certain architectural choices; (2) evaluating the generalization capability of our model on a considerably larger and more heterogeneous data set via Dice coefficient, compactness and Hausdorff distance under different training settings (training set size and different ground truth); and (3) making RatLesNetv2 publicly available.…”

Section: Introductionmentioning

confidence: 56%

“…Manual segmentation can be prohibitively time-consuming as studies involving animals may acquire hundreds of three-dimensional (3D) images. Furthermore, the difficulty of defining lesion boundaries leads to moderate inter- and intra-rater agreement; previous studies have reported that Dice coefficients (Dice, 1945 ) between annotations made by two humans can be as low as 0.73 (Valverde et al, 2019 ) or 0.79 (Mulder et al, 2017a ). Moderate inter-rater agreement is caused by several factors that affect the segmentation quality, including partial volume effect, image contrast and annotator's knowledge and experience.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

RatLesNetv2: A Fully Convolutional Network for Rodent Brain Lesion Segmentation

Valverde

Shatillo²,

Feo

et al. 2020

Front. Neurosci.

Self Cite

View full text Add to dashboard Cite

We present a fully convolutional neural network (ConvNet), named RatLesNetv2, for segmenting lesions in rodent magnetic resonance (MR) brain images. RatLesNetv2 architecture resembles an autoencoder and it incorporates residual blocks that facilitate its optimization. RatLesNetv2 is trained end to end on three-dimensional images and it requires no preprocessing. We evaluated RatLesNetv2 on an exceptionally large dataset composed of 916 T2-weighted rat brain MRI scans of 671 rats at nine different lesion stages that were used to study focal cerebral ischemia for drug development. In addition, we compared its performance with three other ConvNets specifically designed for medical image segmentation. RatLesNetv2 obtained similar to higher Dice coefficient values than the other ConvNets and it produced much more realistic and compact segmentations with notably fewer holes and lower Hausdorff distance. The Dice scores of RatLesNetv2 segmentations also exceeded inter-rater agreement of manual segmentations. In conclusion, RatLesNetv2 could be used for automated lesion segmentation, reducing human workload and improving reproducibility. RatLesNetv2 is publicly available at https://github.com/jmlipman/RatLesNetv2.

show abstract

Section: Introductionmentioning

confidence: 56%

Section: Introductionmentioning

confidence: 99%

RatLesNetv2: A Fully Convolutional Network for Rodent Brain Lesion Segmentation

Valverde

Shatillo²,

Feo

et al. 2020

Front. Neurosci.

Self Cite

View full text Add to dashboard Cite

show abstract

“…This architecture has been shown to generalize even from a limited amount of annotated data [25], and as such is well suited for medical imaging, where datasets as large as the ones commonly used for CNNs are rare. Valverde et al [26] recently demonstrated the effectiveness of U-Net-like architectures in preclinical research, designing the first DNN for the segmentation of ischemic lesions in rodents and achieving segmentation accuracy comparable or better to inter-rater agreement in manual segmentation.…”

mentioning

confidence: 99%

Automated joint skull-stripping and segmentation with Multi-Task U-Net in large mouse brain MRI databases

Feo

Shatillo²,

Sierra

et al. 2020

Preprint

Self Cite

View full text Add to dashboard Cite

11Skull-stripping and region segmentation are fundamental steps in preclinical magnetic resonance imaging (MRI) studies, and these common procedures are usually performed manually. We present Multi-task U-Net (MU-Net), a convolutional neural network designed to accomplish both tasks simultaneously. MU-Net achieved higher segmentation accuracy than state-of-theart multi-atlas segmentation methods with an inference time of 0.35 seconds and no pre-processing requirements. We evaluated the performance of our network in the presence of skip connections and recently proposed framing connections, finding the simplest network to be the most effective. We tested MU-Net with an unusually large dataset combining several independent studies consisting of 1,782 mouse brain MRI volumes of both healthy and Huntington animals, and measured average Dice scores of 0.906 (striati), 0.937 (cortex), and 0.978 (brain mask). These high evaluation scores demonstrate that MU-Net is a powerful tool for segmentation and skull-stripping, decreasing inter and intra-rater variability of manual segmentation. The MU-Net code and the trained model are publicly available at https://github.com/Hierakonpolis/MU-Net. 12 1 Introduction 13 Preclinical imaging studies serve a fundamental role in biological and medical research, relating research results at the 14 molecular level to clinical application in diagnosis and therapy. Magnetic Resonance Imaging (MRI) represents about 23% of 15 all small-animal imaging studies providing the opportunity to monitor the development of pathological condition and response 16to treatment in a non-invasive way 1 . Its unique qualities also include the availability of different imaging contrasts, making MRI 17 especially useful in the context of preclinical neuroscience, from drug development 2 to structural, parametric and functional 18 studies 3 . 19 Skull-stripping and region segmentation represent an integral part of processing pipelines in both human and murine MR 20 imaging. Skull-stripping refers to the identification of the brain within the MRI volume and region segmentation refers to 21 the labeling of specific anatomical regions of interest (ROIs) within the brain. Especially in preclinical imaging, these tasks 22 are often performed manually. While manual segmentation represents the gold standard and is employed as the ground truth 23 when evaluating automated segmentation algorithms, it is a slow process that depends on the expertise of the individual expert. 24 Furthermore, manual segmentation suffers from both intra-and inter-rater variability. Dice scores are typically used to quantify 25 segmentation agreement 4 . Inter-and intra-rater Dice scores reported in literature range between 0.80 to 0.96 across different 26 segmentation tasks, and can be as low as 0.75 in the presence of pathological states 5-7 . 27In preclinical MRI, state-of-the-art automated region segmentation pipelines are based on atlas registration: individual MRI 28 volumes are aligned with a labeled template (atlas) to propagate...

show abstract

“…Some CNN-based networks without pre- or post-processing had been proposed for automatic segmentation of brain lesions on MR images ( 22 , 23 ). However, a pre-processing was still conducted in the present study to normalize all the slices and discard the slices without white matter, gray matter and cerebrospinal fluid (CSF).…”

Section: Methodsmentioning

confidence: 99%

“…The contours of WML and AIL manually delineated by the observers were considered as the gold standard. Subsequently, a rating scheme from 0 to 9 was used to evaluate the WML burden of each individual (21) (Figure 1). Table 2 exhibits the distributions of the WML rating scores in training and testing sets.…”

Section: Gold Standardmentioning

confidence: 99%

Identification of White Matter Lesions in Patients With Acute Ischemic Lesions Using U-net

Liu

et al. 2020

Front. Neurol.

View full text Add to dashboard Cite

Background: White matter lesions (WML) have been proved to be significantly associated with many brain diseases. Precise evaluation of burden of WML at early stage could provide insights in the prognosis and assist in intervention. However, acute ischemic lesions (AIL) exhibit hyperintensities on FLAIR images either, and are detected by diffusion weighted imaging (DWI). It is challenging to identify and segment WML in the patients with WML and AIL. Convolutional neural network (CNN) based architecture has been validated as an efficient tool for automatic segmentation. This study aimed to evaluate the performance of U-net in evaluation of WML in the patients with WML and AIL. Methods: A total of 208 cases from Chinese Atherosclerosis Risk Evaluation (CARE II) study were recruited in the present study. All subjects underwent imaging of FLAIR and DWI on 3.0 Tesla scanners. The contours of WML delineated by the observer and its scores rated by the observer were considered as gold standard. Among all 208 cases, 108 were randomly selected as train set, and the remaining 100 cases were used as test set. The performance of lesion segmentation toolbox (LST) and three U-net models were evaluated on three levels: pixel, lesion, and subject levels. The performance of all methods in WML identification and segmentation was also evaluated among the cases with different lesion volumes and between the cases with and without AIL. Results: All U-net models outperformed LST on pixel, lesion, and subject levels, while no differences were found among three U-net models. All segmentation methods performed best in the cases with WML volume (WMLV) > 20 ml but worst in those with WMLV < 5 ml. In addition, all methods showed similar performance between the cases with and without AIL. The scores determined by U-net exhibited a strong correlation with the gold standard (all Spearman correlation coefficients >0.89, ICCs >0.88, p-values <0.001). Conclusion: U-net performs well on identification and segmentation of WML in the patients with WML and AIL. The performance of U-net is validated by a dataset of multicenter study. Our results indicate that U-net has an advantage in assessing the burden of WML in the patients suffered from both WML and AIL.

show abstract

Automatic Rodent Brain MRI Lesion Segmentation with Fully Convolutional Networks

Cited by 8 publications

References 17 publications

RatLesNetv2: A Fully Convolutional Network for Rodent Brain Lesion Segmentation

RatLesNetv2: A Fully Convolutional Network for Rodent Brain Lesion Segmentation

Automated joint skull-stripping and segmentation with Multi-Task U-Net in large mouse brain MRI databases

Identification of White Matter Lesions in Patients With Acute Ischemic Lesions Using U-net

Contact Info

Product

Resources

About