Dynamic Multi-scale CNN Forest Learning for Automatic Cervical Cancer Segmentation

Bnouni, Nesrine; Rekik, Islem; Rhim, Mohamed Salah; Amara, Najoua Essoukri Ben

doi:10.1007/978-3-030-00919-9_3

Cited by 12 publications

(10 citation statements)

References 16 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Second, we note that the proliferation and view alignment steps are performed in a sequential manner. In the spirit of bi-directional learning introduced in recent medical image segmentation works (Amiri et al, 2018;Bnouni et al, 2018) and which has outperformed sequential learning, we can design a proliferation-alignment bidirectional learning model, where the proliferated data improves the learning of the joint mapping transformation, and in turn the learned multi-view data mapping improves the quality of the proliferated data. This is a novel research direction that we intend to investigate in our future work building on this seminal model.…”

Section: Discussionmentioning

confidence: 99%

Multi-view learning-based data proliferator for boosting classification using highly imbalanced classes

Graa

Rekik

2019

Journal of Neuroscience Methods

Self Cite

View full text Add to dashboard Cite

Background. Multi-view data representation learning explores the relationship between the views and provides rich complementary information that can improve computer-aided diagnosis. Specifically, existing machine learning methods devised to automate neurological disorder diagnosis using brain data provided new insights into how a particular disorder such as autism spectrum disorder (ASD) alters the brain construct. However, the performance of machine learning methods highly depends on the size of the training samples from both classes. In a real-world clinical setting, such medical data is very expensive and challenging to collect, might (i) su↵er from several limitations such as imbalanced classes and (ii) have non-heterogeneous distribution when derived from multi-view brain representations.-New Method. To the best of our knowledge, the problem of imbalanced and multi-view data classification remains unexplored in the field of network neuroscience. To fill this gap, we propose a Multi-View LEArning-based data Proliferator (MV-LEAP) that enables the classification of imbalanced multiview representations. MV-LEAP comprises two key steps. First, a manifold learning-based proliferator, which enables to generate synthetic data for each view, is developed to handle imbalanced data. Second, a multi-view manifold data alignment leveraging tensor canonical correlation analysis is proposed to map all original and proliferated (i.e., synthesized) views into a shared subspace where their distributions are aligned for the target classification task.

show abstract

Section: Discussionmentioning

confidence: 99%

Multi-view learning-based data proliferator for boosting classification using highly imbalanced classes

Graa

Rekik

2019

Journal of Neuroscience Methods

Self Cite

View full text Add to dashboard Cite

show abstract

“…Nevertheless, the PLN appearance will be helpful in predicting the node status when utilized with other criteria. We calculate the border values with the energy (E) measure from Gray-Level Co-Occurrence Matrix (GLCM) according to (2). Here we use a RG algorithm for LN segmentation, but other semiautomatic methods can be used.…”

Section: Ratio = (Longaxis)/(shortaxis)mentioning

confidence: 99%

“…Endometrial and cervical cancer are the mostly common gynecologic malignancies in the world. The clinical stage [1] is based on the prognostic factors like the tumor volume [2] and the nodal status. Cancer can spread to other part of the body and/or through the lymph system.…”

Section: Introductionmentioning

confidence: 99%

Computer-aided Lymph Node Detection using Pelvic Magnetic Resonance Imaging

Bnouni

Rekik

Rhim

et al. 2020

IJCDS

View full text Add to dashboard Cite

Pelvic Lymph Nodes (PLNs) segmentation and classification are fundamental tools in the medical image analysis of pelvic gynecological cancer such as endometrial and cervical cancer. Often used by the radiologist, PLN classification requires detailed knowledge of the morphological features of PLNs, derived from size, shape, contour and heterogeneous appearance. Accurate PLN segmentation is an essential step in PLN classification. In order to supply the best assessment of a nodal status, semiautomatic and automatic PLN segmentation and classification methods are highly desired as they can strongly capture the wide variability in morphological features and reduce classification errors due to the inter and intra-observer variability, while avoiding the time-consuming for manual delineation of PLN boundary. Nevertheless, semi-automatic segmentation methods require the clinician intervention to select the initial seed point. However, typical semi-automatic PLN segmentation methods might fail due to (1) the intensity inhomogeneity, noise and low contrast in medical images, and (2) the position of the starting point. Thus, the performance of these methods can be enhanced by using a preprocessing-based iterative segmentation approach. Currently, Magnetic Resonance Imaging (MRI) is the most common imaging modality used for staging endometrial and cervical cancer, evaluation of PLN involvement and selection of therapeutic strategy. PLN detection using classic features can be challenging due to the similarity between normal and abnormal PLNs structures. In pelvic cancer and metastatic PLN, Diffusion Weighted (DW)-MRI exhibits brighter areas indicating tumor and metastatic PLN. This paper combines anatomic T2-Weighted (T2-w) imaging with DW imaging. Specifically, we propose a computer-aided pelvic framework, which leverages (1) an ensemble preprocessing method to improve PLN segmentation, (2) the iterative correction of the position of the initial point by executing the segmentation algorithm several times in succession, (3) the fusion of structural and diffusion MRI and, (4) the extraction of morphological features of segmented PLNs (axial T2-w image) as well as intensity feature derived from the fused image for the final classification of PLNs as suspect or non-suspect. Research in the field of PLN detection is important as it can help doctors to better detect cervical and endometrial cancer and decide the appropriate treatment. To the best of our knowledge, this is the first work to segment and classify PLN. Our preprocessing-based iterative segmentation approach significantly (p<0.05) improved comparison segmentation methods, with a segmentation accuracy boosted from 61.37% for the conventional region-growing algorithm to 66.53% for the proposed method. Furthermore, we obtained an average accuracy of 78.50% for pelvic nodule classification.

show abstract

“…Second, so far, we have only considered a unidirectional interaction from a source view to a target view given that the target view holds more relevant information for the classification in hand. To generalize our framework in the case where the views hold equally relevant information that is complementary, we can define a bidirectional flow across views in the spirit of [6], [42] where different classifiers are trained to exchange bidirectional information at multiple scales to improve learning. Third, up to this point, our framework can only handle two views.…”

Section: E Limitations and Future Workmentioning

confidence: 99%

“…T2w-MRI images of the pelvis are acquired in axial, sagittal and coronal planes. In cervical cancer, while axial-T2w images arranged perpendicular to the long axis of cervix yield more precise evaluation of the parametrial invasion, the nodes status [5] and the stromal involvement, sagittal-T2w images planned parallel to the long axis of cervix provide more accurate assessment of the tumor size [6], [7] and the extension of neighboring organs.…”

Section: Introductionmentioning

confidence: 99%

Cross-View Self-Similarity Using Shared Dictionary Learning for Cervical Cancer Staging

et al. 2019

Self Cite

View full text Add to dashboard Cite

Dictionary Learning (DL) has gained large popularity in solving different computer vision and medical image problems. However, to the best of our knowledge, it has not been used for cervical tumor staging. More importantly, there have been very limited works on how to aggregate different interactions across data views using DL. As a contribution, we propose a novel cross-view self-similarity low rank shared dictionary learning-based (CVSS-LRSDL) framework, which introduces three major contributions in medical image-based cervical cancer staging: (1) leveraging the complementary of axial and sagittal T2w-magnetic resonance (MR) views for cervical cancer diagnosis, (2) introducing self-similarity (SS) patches for DL training, which explore the unidirectional interaction from a source view to a target one, and (3) extracting features that are shared across tumor grades and grade-specific features using the CVSS-LRSDL learning approach. For the first and second contributions, given an input patch in the source view (axial T2w-MR images), we generate its SS patches within a fixed neighborhood in the target view (sagittal T2w-MR images). Specifically, we produce a unidirectional patch-wise SS from a source to a target view, based on mutual and additional information between both views. As for the third contribution, we represent each individual subject using the weighted distance matrix between views, which is used to train our DL-based classifier to output the label for a new testing subject. Overall, our framework outperformed several DL based multi-label classification methods trained using: (i) patch intensities, (ii) SS single-view patches, and (iii) weighted-SS single-view patches. We evaluated our CVSS-LRSDL framework using 15 T2w-MRI sequences with axial and sagittal views. Our CVSS-LRSDL significantly (p<0.05) outperformed several comparison methods and obtained an average accuracy of 81.73% for cervical cancer staging. INDEX TERMS Shared dictionary learning, low-rank models, self-similarity, cross-view, cervical cancer stage.

show abstract

Dynamic Multi-scale CNN Forest Learning for Automatic Cervical Cancer Segmentation

Cited by 12 publications

References 16 publications

Multi-view learning-based data proliferator for boosting classification using highly imbalanced classes

Multi-view learning-based data proliferator for boosting classification using highly imbalanced classes

Computer-aided Lymph Node Detection using Pelvic Magnetic Resonance Imaging

Cross-View Self-Similarity Using Shared Dictionary Learning for Cervical Cancer Staging

Contact Info

Product

Resources

About