Improving Generalizability in Polyp Segmentation using Ensemble Convolutional Neural Network

Polyp segmentation is crucial for the diagnosis of colorectal cancer. Early detection and removal of polyps can prolong the life of patients and reduce the mortality rate. Despite near expert-label performance with applying the deep learning method in polyp segmentation tasks, the generalization of such models in the clinical environment remains a significant challenge. Transfer learning from a large medical dataset from the same domain is a common technique to address generalizability. However, it is difficult to find a similar large medical dataset. In this work, we investigate the feasibility of building a generalizable model for polyp segmentation using an ensemble of four MultiResUNet architectures, each trained on the combination of the different centered datasets provided by the challenge organizers. Our method achieved a decent performance of 0.6172 ± 0.0778 for the multi-centered dataset. Our findings show that significant work needs to be done to design a robust segmentation model for the development of a clinically acceptable system.

[1]  M. Riegler,et al.  PolypGen: A multi-center polyp detection and segmentation dataset for generalisability assessment , 2021, ArXiv.

[2]  Irina Voiculescu,et al.  Deep learning for detection and segmentation of artefact and disease instances in gastrointestinal endoscopy , 2021, Medical Image Anal..

[3]  A. Jemal,et al.  Global Cancer Statistics 2020: GLOBOCAN Estimates of Incidence and Mortality Worldwide for 36 Cancers in 185 Countries , 2021, CA: a cancer journal for clinicians.

[4]  Yining Wang,et al.  Domain Adaptation for Medical Image Segmentation: A Meta-Learning Method , 2021, J. Imaging.

[5]  Michael Riegler,et al.  A Comprehensive Study on Colorectal Polyp Segmentation With ResUNet++, Conditional Random Field and Test-Time Augmentation , 2021, IEEE Journal of Biomedical and Health Informatics.

[6]  Klaus H. Maier-Hein,et al.  Comparative validation of multi-instance instrument segmentation in endoscopy: Results of the ROBUST-MIS 2019 challenge , 2020, Medical Image Anal..

[7]  Haavard D. Johansen,et al.  Real-Time Polyp Detection, Localisation and Segmentation in Colonoscopy Using Deep Learning , 2020, ArXiv.

[8]  Sharib Ali,et al.  DDANet: Dual Decoder Attention Network for Automatic Polyp Segmentation , 2020, ICPR Workshops.

[9]  Bogdan J. Matuszewski,et al.  Polyp Segmentation with Fully Convolutional Deep Neural Networks—Extended Evaluation Study , 2020, J. Imaging.

[10]  Michael A. Riegler,et al.  DoubleU-Net: A Deep Convolutional Neural Network for Medical Image Segmentation , 2020, 2020 IEEE 33rd International Symposium on Computer-Based Medical Systems (CBMS).

[11]  Namkug Kim,et al.  Real-time detection of colon polyps during colonoscopy using deep learning: systematic validation with four independent datasets , 2020, Scientific Reports.

[12]  Felix Y Zhou,et al.  An objective comparison of detection and segmentation algorithms for artefacts in clinical endoscopy , 2020, Scientific Reports.

[13]  Thomas de Lange,et al.  Kvasir-SEG: A Segmented Polyp Dataset , 2019, MMM.

[14]  Mohammad Sohel Rahman,et al.  MultiResUNet : Rethinking the U-Net Architecture for Multimodal Biomedical Image Segmentation , 2019, Neural Networks.

[15]  Thomas de Lange,et al.  ResUNet++: An Advanced Architecture for Medical Image Segmentation , 2019, 2019 IEEE International Symposium on Multimedia (ISM).

[16]  T. Berzin,et al.  Real-time automatic detection system increases colonoscopic polyp and adenoma detection rates: a prospective randomised controlled study , 2019, Gut.

[17]  Osamu Abe,et al.  Deep learning and artificial intelligence in radiology: Current applications and future directions , 2018, PLoS medicine.

[18]  T. Berzin,et al.  Development and validation of a deep-learning algorithm for the detection of polyps during colonoscopy , 2018, Nature Biomedical Engineering.

[19]  Hugo Larochelle,et al.  Optimization as a Model for Few-Shot Learning , 2016, ICLR.

[20]  Fernando Vilariño,et al.  WM-DOVA maps for accurate polyp highlighting in colonoscopy: Validation vs. saliency maps from physicians , 2015, Comput. Medical Imaging Graph..

[21]  Thomas Brox,et al.  U-Net: Convolutional Networks for Biomedical Image Segmentation , 2015, MICCAI.

[22]  Dumitru Erhan,et al.  Going deeper with convolutions , 2014, 2015 IEEE Conference on Computer Vision and Pattern Recognition (CVPR).