Automated Polyp Segmentation in Colonoscopy using MSRFNet

Abstract

Colorectal cancer is one of the major cause of cancer-related death around the world. High-quality colonoscopy is considered mandatory for resecting and preventing colorectal cancers. In the recent past, various technological advances have been made towards improving the quality of colonoscopy. Despite the technical advancement, some polyps are frequently missed during colonoscopy examinations. Polyp detection ( for example, adenomas) rates are largely influenced by inter-endoscopist variability. Therefore, it is very challenging to standardize a high-quality colonoscopy. A computeraided detection system could solve the problem with miss-detection. The “MediaEval 2021” challenge entails the chance to study and develop accurate automated polyp segmentation algorithms [6]. In this paper, we propose our approach based on MSRFNet. Our experimental findings show that the model trained on the KvasirSEG dataset and evaluated on a competition test dataset obtains a dice coeficient of 0.7055, Jaccard of 0.6176, a recall of 0.7293, and a precision of 0.7769. In addition to the MediaEval 2021 challenge, we evaluated our approach on the Endotect Challenge Dataset and “2020 Medico Automatic Polyp Segmentation Challenge Dataset". The results further demonstrate the eficiency of our approach.

INTRODUCTION

Colorectal cancer is the third leading cause of cancer-related death globally [ 12 ]. Although colonoscopy has improved the detection of colorectal polyps, computer-aided detection could better indicate the presence and location of polyps in the colon. A CAD system could assist endoscopists by finding out the missed polyps. One of the other significant advantages of the CAD system is that they are not influenced human bias or inter and intraobserver variability. Therefore, such systems could improve clinical performance irrespective of gastroenterologists expertise. In this respect, we propose our approach based on MSRFNet [ 13 ], which was specially designed for the segmentation of medical images.

In summary, the main contribution of the paper are as follows: 2 Automated polyp segmentation is a well-established topic. There has been several works proposed on the automated polyp segmentation [ 1, 4, 9, 16 ]. In addition to the individual work, there are several competitions and challenges held in order to solve the polyp segmentation problem [ 1, 2, 5, 7, 8 ]. Paudel et al.[ 10 ] provided a winning solution for the 2020 “Medico automatic polyp segmentation challenge”. They used eficientNet [ 14 ] as an encoder and UNet [ 11 ] as decoder. Additionally, they also used a channel-spatial attention module and deep supervision to improve the performance of the network. Similarly, Thambawita et al. [ 15 ] provided another solution for the segmentation task at the 3rd International Endoscopy Computer Vision Challenge and Workshop (EndoCV2021). The proposed architecture, DivergentNets, combined TriUNet with UNet++, FPN, DeepLabv3, and DeepLabv3+, into a single model to achieve generalizable performance. “Medico: Transparency in Medical Image Segmentation" challenge aims to develop transparent and explainable automated methods. We participate in this challenge to provide our efective solution and benchmark our solutions against other participants on the same test dataset. 3

DATASET

HyperKvasir [ 3 ] was provided by the challenge organizers as the development dataset. Kvasir-SEG consists of 1000 images, their corresponding ground truth and the bounding box information. As the test dataset, the organizers provided us with 200 unique images consisting of at least one polyp image. Similarly, we performed further experiments on the Endotect Challenge Dataset and Medico Automatic Polyp Segmentation Challenge Dataset. 1https://endotect.com/ 2https://multimediaeval.github.io/editions/2020/tasks/medico/

METHODOLOGY

MSRFNet architecture and its components are shown in Figure. 1. MSRFNet has a dual-scale dense fusion (DSDF) block that consists of residual dense connections and is capable of transferring data across diferent scales. It is a fully convolutional network that computes the multi-scale features and fuses them efectively using a DSDF block. The residual nature of the DSDF block improves gradient flow which improves the training eficiency. Due to page limitations, for more details on the working, components and architecture details of MSRFNet, readers are requested to refer [ 13 ]. 5

RESULTS AND ANALYSIS

It can be observed from Table 1 that our trained MSRFNet model achieved a dice coeficient of 0.7055, Jaccard index of 0.6167, and a precision of 0.7769 on the MediaEval organiser’s test dataset, manifesting MSRFNet generalization capabilities. Similarly, the qualitative results are demonstrated in Figure2. The qualitative results show both obvious polyps and dificult polyps. From the qualitative results, it can be observed that our approach is able to detect the obvious polyps, including small polyps, however, it fails in challenging cases. Due to the unavailability of the ground truth, we can only present the original images and the predicted masks.

From the Table 2, we can observe that our model obtains descent performance for both datasets that further demonstrates the eficiency of our approach. 6

CONCLUSION & FUTURE WORK

We competed on the organizer’s dataset using MSRFNet architecture and achieved a dice coeficient of 0.7055, Jaccard index of 0.6176, a recall of 0.7293, and a precision of 0.7769. In the polyp segmentation challenge task, the MSRFNet performed well, as depicted by diferent performance metrics. We further evaluated our results on the Endotect Challenge Dataset and 2020 Medico Automatic Polyp Segmentation Challenge Dataset that demonstrated the eficiency of our approach.

In the future, we want to enhance the MSRFNet performance design by assessing the best hyperparameter settings for the automatic polyp segmentation.

ACKNOWLEDGMENTS

The computations in this paper were performed on the equipment provided by the Experimental Infrastructure for Exploration of Exascale Computing (eX3), which is financially supported by the Research Council of Norway under the contract 270053. Medico: Transparency in Medical Image Segmentation

References

[1] Sharib Ali , Felix Zhou, Barbara Braden, Adam Bailey, Suhui Yang, Guanju Cheng, Pengyi Zhang, Xiaoqiong Li, Maxime Kayser , Roger D Soberanis-Mukul , and others. 2020 . An objective comparison of detection and segmentation algorithms for artefacts in clinical endoscopy . Scientific reports 10 , 1 ( 2020 ), 1 - 15 .

[2] Jorge Bernal , Nima Tajkbaksh, Francisco Javier Sanchez, Bogdan J Matuszewski, Hao Chen, Lequan Yu, Quentin Angermann, Olivier Romain, Bjørn Rustad, Ilangko Balasingham, and others. 2017 . Comparative validation of polyp detection methods in video colonoscopy: results from the MICCAI 2015 endoscopic vision challenge . IEEE transactions on medical imaging 36 , 6 ( 2017 ), 1231 - 1249 .

[3] Hanna Borgli , Vajira Thambawita, Pia H Smedsrud, Steven Hicks, Debesh Jha, Sigrun L Eskeland, Kristin Ranheim Randel, Konstantin Pogorelov, Mathias Lux, Duc Tien Dang Nguyen, Dag Johansen, Carsten Griwodz, Håkon K Stensland, Enrique Garcia-Ceja, Peter T Schmidt, Hugo L Hammer, Michael A Riegler, Pål Halvorsen , and Thomas de Lange. 2020 . HyperKvasir, a comprehensive multi-class image and video dataset for gastrointestinal endoscopy . Scientific Data 7 , 1 ( 2020 ), 283 . https://doi.org/10.1038/s41597-020-00622-y

[4] Yun Bo Guo and Bogdan Matuszewski . 2019 . Giana polyp segmentation with fully convolutional dilation neural networks . In Proceedings of the 14th International Joint Conference on Computer Vision , Imaging and Computer Graphics Theory and Applications . 632 - 641 .

[5] Steven Hicks , Debesh Jha, Vajira Thambawita, Pål Halvorsen, BjørnJostein Singstad, Sachin Gaur, Klas Pettersen, Morten Goodwin, Sravanthi Parasa, Thomas de Lange, and others. 2021 . MedAI: Transparency in Medical Image Segmentation . Nordic Machine Intelligence 1 , 1 ( 2021 ), 1 - 4 .

[6] Steven Hicks , Debesh Jha, Vajira Thambawita, Hugo Hammer, Thomas de Lange, Sravanthi Parasa, Michael Riegler , and Pål Halvorsen . 2021 . Medico Multimedia Task at MediaEval 2021: Transparency in Medical Image Segmentation . In Proceedings of MediaEval 2021 CEUR Workshop.

[7] Steven A Hicks , Debesh Jha, Vajira Thambawita, Pål Halvorsen, Hugo L Hammer, and Michael A Riegler. 2021 . The EndoTect 2020 challenge: evaluation and comparison of classification, segmentation and inference time for endoscopy . In International Conference on Pattern Recognition . 263 - 274 .

[8] Debesh Jha , Steven A Hicks, Krister Emanuelsen, Håvard Johansen, Dag Johansen, Thomas de Lange, Michael A Riegler, and Pål Halvorsen . 2020 . Medico multimedia task at mediaeval 2020: Automatic polyp segmentation . arXiv preprint arXiv: 2012 . 15244 ( 2020 ).

[9] Ge-Peng Ji , Yu-Cheng Chou , Deng-Ping Fan , Geng Chen, Huazhu Fu, Debesh Jha, and Ling Shao . 2021 . Progressively Normalized SelfAttention Network for Video Polyp Segmentation. In In proceedings of Medical image computing and computer assisted intervention .

[10] Sahadev Poudel and Sang-Woong Lee . 2020 . Automatic Polyp Segmentation Using Channel-Spatial Attention with Deep Supervision . ( 2020 ).

[11] Olaf Ronneberger , Philipp Fischer , and Thomas Brox . 2015 . U-net: Convolutional networks for biomedical image segmentation . In International Conference on Medical image computing and computer-assisted intervention . 234 - 241 .

[12] Juan Silva , Aymeric Histace, Olivier Romain, Xavier Dray, and Bertrand Granado . 2014 . Toward embedded detection of polyps in wce images for early diagnosis of colorectal cancer . International journal of computer assisted radiology and surgery 9 , 2 ( 2014 ), 283 - 293 .

[13] Abhishek Srivastava , Debesh Jha, Sukalpa Chanda, Umapada Pal, Hå- vard D Johansen , Dag Johansen , Michael A Riegler, Sharib Ali , and Pål Halvorsen . 2021 . MSRF-Net: A Multi-Scale Residual Fusion Network for Biomedical Image Segmentation . arXiv preprint arXiv:2105.07451 ( 2021 ).

[14] Mingxing Tan and Quoc Le . 2019 . Eficientnet: Rethinking model scaling for convolutional neural networks . In International Conference on Machine Learning . 6105 - 6114 .

[15] Vajira Thambawita , Steven A Hicks, Pål Halvorsen, and Michael A Riegler . 2021 . Divergentnets: Medical image segmentation by network ensemble . arXiv preprint arXiv:2107.00283 ( 2021 ).

[16] Zongwei Zhou , Md Mahfuzur Rahman Siddiquee, Nima Tajbakhsh, and Jianming Liang . 2019 . Unet++: Redesigning skip connections to exploit multiscale features in image segmentation . IEEE transactions on medical imaging 39 , 6 ( 2019 ), 1856 - 1867 .