Improving automated multiple sclerosis lesion segmentation with a cascaded 3D convolutional neural network approach

By Sergi Valverde, Mariano Cabezas, E. Roura, Sandra González-Villà, D. Pareto, Joan Carles Vilanova, L. Ramió-Torrentá, À. Rovira, Arnau Oliver, Xavier Lladó. NeuroImage, 2017

391 citations recorded by the source corpus.

ABSTRACT In this paper, we present a novel automated method for White Matter (WM) lesion segmentation of Multiple Sclerosis (MS) patient images. Our approach is based on a cascade of two 3D patch‐wise convolutional neural networks (CNN). The first network is trained to be more sensitive revealing possible candidate lesion voxels while the second network is trained to reduce the number of misclassified voxels coming from the first network. This cascaded CNN architecture tends to learn well from a small (Symbol) set of labeled data of the same MRI contrast, which can be very interesting in practice, given the difficulty to obtain manual label annotations and the large amount of available unlabeled Magnetic Resonance Imaging (MRI) data. We evaluate the accuracy of the proposed method on the public MS lesion segmentation challenge MICCAI2008 dataset, comparing it with respect to other state‐of‐the‐art MS lesion segmentation tools. Furthermore, the proposed method is also evaluated on two private MS clinical datasets, where the performance of our method is also compared with different recent public available state‐of‐the‐art MS lesion segmentation methods. At the time of writing this paper, our method is the best ranked approach on the MICCAI2008 challenge, outperforming the rest of 60 participant methods when using all the available input modalities (T1‐w, T2‐w and FLAIR), while still in the top‐rank (3rd position) when using only T1‐w and FLAIR modalities. On clinical MS data, our approach exhibits a significant increase in the accuracy segmenting of WM lesions when compared with the rest of evaluated methods, highly correlating (Symbol) also with the expected lesion volume. Symbol. No caption available. Symbol. No caption available. HIGHLIGHTSWe propose an automated WM lesion segmentation method for MS patient images.The approach relies on a cascade of two 7‐layer convolutional neural networks.We evaluate its accuracy with both the MICCAI2008 challenge and clinical MS data.Our approach is currently the best ranked method of the challenge (1th pos / 60).On MS data, the accuracy is significantly better that state‐of‐the‐art methods.

Subfields: Computer Vision and Pattern Recognition, Epidemiology, Pathology and Forensic Medicine

View Paper: https://www.semanticscholar.org/paper/714d4502529ef6433c9f5f58ce2337cbac1dbb6f

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Vision and Pattern Recognition

In Epidemiology

Citations in the corpus

Deep convolutional neural networks for brain image analysis on magnetic resonance imaging: a review (2017, methodology): “[124] presented a cascaded 3D CNN approach for addressing MS lesion segmentation.”
Evaluation of a deep learning approach for the segmentation of brain tissues and white matter hyperintensities of presumed vascular origin in MRI (2017, neutral): “Valverde et al. (2017) presented a method that uses a sequence of two CNNs where the first network is used for an initial segmentation and the second network is used to finetune the segmentations.”
Tversky Loss Function for Image Segmentation Using 3D Fully Convolutional Deep Networks (2017, methodology): “A common approach to account for data imbalance, especially in voxelwise methods, is to extract equal training samples from each class [20].”
Standardized Assessment of Automatic Segmentation of White Matter Hyperintensities; Results of the WMH Segmentation Challenge (2019, methodology): “nic-vicorob a 10-layer 3D convolutional neural network architecture previously used to segment multiple sclerosis lesions [42].”
Boundary loss for highly unbalanced segmentation (2018, methodology): “In medical image segmentation, a common strategy is to re-balance class prior distributions by down-sampling frequent labels (Havaei et al., 2017; Valverde et al., 2017).”
HyperDense-Net: A Hyper-Densely Connected CNN for Multi-Modal Image Segmentation (2018, background): “They achieved state-of-the-art performances in a broad range of medical image segmentation problems [26], [27], including multimodal tasks [4]–[6], [15]–[17], [19], [22], [23], [28], [29].”
Deep learning in radiology: an overview of the concepts and a survey of the state of the art with focus on MRI (2018, background): “…approach has found usage in diﬀerent types of segmentation tasks in MRI, for example brain tumor segmentation in [57, 58, 59], white matter segmentation in multiple sclerosis patients [60], segmentation of 25 diﬀerent structures in brain [61], and for rectal cancer segmentation in pelvis MRI [62].”
Unpaired Multi-Modal Segmentation via Knowledge Distillation (2020, background): “This strategy has demonstrated effectiveness on segmenting brain tissue [3]–[5] and brain lesions [6]–[8] in multiple sequences of MRI.”
Dice Loss for Data-imbalanced NLP Tasks (2019, methodology): “Empirically, these two methods are not widely used due to the trickiness of selecting α especially for multi-class classification tasks and that inappropriate selection can easily bias towards rare classes (Valverde et al., 2017).”
Deep learning based brain tumor segmentation: a survey (2020, neutral): “CNN Single-Path [26], [27], [28], [29], [30], [31], [32] Multi-Path [33], [34], [35], [36], [37], [38], [39], [40], [41], [42] FCN Vanilla FCN [43], [44], [28], [45], [46], [47], [48], [49], [50] U/V-Net Based [51], [52], [53], [54], [55], [56], [57], [58], [59], [60] Cascaded CNN [61], [62], [63], ”