GastroEffNetV1- CNN based Automated detection of Gastrointestinal abnormalities from capsule endoscopy images

doi:10.21203/rs.3.rs-2588671/v1

Download PDF

Research Article

GastroEffNetV1- CNN based Automated detection of Gastrointestinal abnormalities from capsule endoscopy images

https://doi.org/10.21203/rs.3.rs-2588671/v1

This work is licensed under a CC BY 4.0 License

Version 1

posted

You are reading this latest preprint version

Purpose: Gastrointestinal disorders are a class of prevalent disorders in the world. Capsule endoscopy is considered as an effective diagnostic modality for diagnosis of such gastrointestinal disorders.

Aim: The work is to leverage an algorithm for automated classification of the gastrointestinal abnormalities using capsule endoscopy images using Deep learning algorithms.

Method: In this method we proposed a deep learning architecture GastroEffNetV1 for automatic classification of the abnormalities in the capsule endoscopy images. The gastrointestinal abnormalities considered are ulcerative colitis, polyps and esophagitis. The curated dataset consists of 6000 images with ground truth labeling. A website was developed using the trained algorithm to execute automatic classification of the input image as either ulcerative colitis, polyp, esophagitis or as normal condition.

Result: The classifier produced 99.15% validation accuracy, 0.0918 validation loss, 99.25% specificity and 99.25% sensitivity and 0.991 AUC. These results exceed that of the state-of-the-art systems.

Conclusion: Hence the GastroEffNetV1 could be used to identify the different gastrointestinal abnormalities in the capsule endoscopy image which will in turn increase quality of healthcare.

Gastrointestinal disorders

Capsule endoscopy

Convolutional Neural Network

Transfer Learning

Deep learning

GastroEffnetV1

The disorders related to the digestive system are called Gastrointestinal disorders. These disorders cause economic and social impact on our society. The most common intestinal disorders are irritable bowel syndrome (IBS), GERD, Ulcer, Polyps, Colon cancer, Gastroenteritis, Diverticular disease, Celiac disease (CD), Inflammatory bowel disease, Gastric cancer, hemorrhoids, Crohn’s disease and pancreatitis. Among these most commonly affected is IBS,the predominance of it within the common population universally is 4.1%. IBS is 1.6 times more frequent in females than in males[1]. Amongst the developing regions of the world, India has the most noteworthy prevalence frequency of 9.31 for IBD and 5.41 for ulcerative colitis per 100000 persons. CD which has a prevalence rate of 1% in various countries is predominantly observed in male sex in childhood but gets to be predominant in women in adulthood, while the rate of UC is reported to be equal between the genders from childhood to adulthood [2].

Ulcerative colitis is an inflammatory bowel disease which causes inflammation on the digestive tract. The disorder begins in the rectum and for the most part extends proximally in a nonstop way through the complete colon. Around 5·7–15·5% of patients with ulcerative colitis have a first-degree relative with the same disease. Ulcerative colitis is more predominant than Crohn’s disease. Patients with longstanding and/or extensive UC have an expanded chance of colorectal cancer (CRC) [4]. Gastric polyp is an unusual development of tissue projecting from the gastric mucosal layer. In general, the rate of polyps shows up to have expanded, as indicated by a higher prevalence in huge series [5]. Evacuating the polyps essentially diminishes the frequency and mortality of Colorectal cancer. Colonoscopy is the favored screening tool because it permits direct examination of the colorectal mucosa and evacuation of polyps with malignant potential [6]. Esophagitis refers to inflammation or injury to the esophageal mucosa. The common cause is gastroesophageal reflux, which can lead to erosive esophagitis. The symptoms are vomiting, regurgitation, nausea, epigastric abdominal pain or chest pain, dysphagia, water brash, globus, or decreased appetite. As numerous as one third of patients with esophagitis may have an ordinary showing up esophagus. Other findings incorporate esophageal wrinkles, strictures, mucosal rings (trachealization), and white Medication-induced esophagitis has an estimated incidence of 3.9 per 100,000 population per year with a mean age at diagnosis of 41.5 years [7].

The traditional imaging and endoscopic procedures have aided clinicians to examine and diagnose the human gastro-intestinal (GI) tract. Using these radiological procedures we can examine polyps, inflammation and punctures in the GI tract. The basic radiology methods like barium swallow and X-ray fluoroscopy are used to visualize and examine the upper gastrointestinal tract [8]. Invasive and non-surgical procedures like Upper endoscopy, Gastroscopy, Colonoscopy and ultra-high magnification Endocytoscopy are used to identify bleeding, density, severity of ulcer, biopsy collection and tumor identification [9]. One problem associated with these methods is that they can cause duodenal hematoma, infection, abdominal pain, or tear in the colon wall during the process of diagnosis. Balloon enteroscopy allow examining the colon polyps or areas of bleeding in the gastrointestinal (GI) tract. Some of the complications associated with this method are sedation, perforation and potent risk of ileus (transient slowing of the bowel) [10]. The disadvantage of conventional endoscopy such as Colonoscopy or Esophagogastroduodenoscopy is the inability to image and examine the small intestine. This could be overcome by using capsule endoscopy in which a capsule is ingested by the patient and the capsule images throughout the entire GI tract. Capsule endoscopy includes limited preparation and no anesthesia, painless procedure and is the best imaging diagnostic modality for gastrointestinal abnormalities [11].

1.1 Related Work

Wang, S., et al proposed a systematic evaluation and optimization of automatic detection of ulcers in wireless capsule endoscopy on a large dataset collected from more than 30 hospitals and 100 medical examination centers using deep convolutional neural networks. A method called second glance (secG) detection framework for automatic detection of ulcers was used [12]. Barash, Y., et al proposed a method to grade the severity of ulcer in video-capsule images of Crohn’s disease patients using an ordinal neural network solution. Deep learning algorithm was used to automate the grading. The dataset used was provided by PIllCamCrohn’s Capsule (PCC), Medtronic [13]. Saito, H., et al developed and tested a convolutional neural network to automatically detect protruding lesions of various types in WCE (Wireless Capsule Endoscopy) images. The data were collected using Pillcam SB2 and were analyzed using STATA. He constructed an AI system using SSD and all layers of the CNN were fine-tuned using stochastic gradient descent [14].

Ghosh, T., et al developed a computer-aided diagnostic (CAD) tool for automated analysis of small intestinal abnormalities like bleeding. A VGG16 network was used for the convolutional layers and a softmax classifier was fed by the decoder output feature maps for pixel-wise classification [15]. Yuan, Y., et al proposed a two-stage fully automated computer-aided detection system to detect ulcer from WCE images by saliency max-pooling method along with Locality-constrained Linear Coding (LLC) Method. In this method,170 ulcer images and 170 normal images were collected and the classification was done using Support Vector Machine (SVM) [16].

2.1 Material Description

KVASIR and ETIS- Larib are certain public datasets available for wireless capsule endoscopy [17, 18]. The dataset used in this work is curated from these sources related to wireless capsule endoscopy. The dataset consists of polyp, ulcerative colitis, esophagitis and normal conditions. The dataset consists of 6000 images for ground truth showing anatomical landmarks, pathological findings or endoscopic procedures in the GI tract. The anatomical landmarks are Z-line, pylorus and cecum, while the pathological finding includes esophagitis, polyps and ulcerative colitis.

Table 1

Image distribution in the WCE dataset
Class No	Class Name	Total Images for Ground Truth
0	Normal	1500
1	Ulcerative colitis	1500
2	Polyps	1500
3	Esophagitis	1500

2.2 Designated Workflow

The proposed workflow is about the automated diagnosis of gastrointestinal abnormalities from capsule endoscopy images using CNN. The image is fetched from the user and suitable image processing technique is applied. The CNN is trained on capsule endoscopy images which will automatically classify the processed image into one of the 4 classes (normal, ulcerative colitis, polyps, esophagitis).

2.3 Architecture for classifier model

In this work we propose a deep learning model, GastroEffNetV1 which is based on EfficientNetV2B2 architecture. Figure 2 represents the architecture diagram of the proposed model used in this work.

The EfficientNet scaling method uniformly scales network width, depth, and resolution with a set of fixed scaling coefficients and the input of one convolution layer is fed as input to another convolution layer appearing later in the network[27]. This connection eliminated the problem of data morphing and using this concept deep CNNs came into existence. This feature has been established in a motive to overcome the expense of convolution operation. It consumes less time than the standard convolution and hence is efficient. Apart from that, there are no non-linearities introduced after the operation of convolution.EfficientNetV2B2 belonging to the EfficientNet family is a convolutional neural network architecture. Apart from EfficientNetV2B2, several other standard CNN architectures like VGG16, VGG19, ResNet101 and other models from the EfficientNet family have been used for comparative analysis.

In addition to standard EfficientNetV2B2 architecture we added some more enhancements like Gaussian Noise, Batch Normalization, Dropout and Gaussian noise is used for near immediate smoothing on tested models. The goal of the filtering action is to cancel noise while preserving the integrity of edge and detail information, nonlinear approaches generally provide more satisfactory results than linear techniques[30].Batch Normalization (BN) and its variations have conveyed huge success in combating the covariate shift induced by the training step of deep learning methods. Whereas these techniques normalize feature distributions by standardizing with batch statistics, they don't correct the influence on features from extraneous variables or multiple distributions [25].Dropout approach that randomly shuts down feature detectors during the training phase. This enables the network to become invariant to the ‘noisy’ spectral components by randomly selecting only the most important basis vectors for signal reconstruction during the spectral dropout regularization process[26].These layers were added for robustness and to overcome the over fitting.

2.4 Data Augmentation

Data augmentation is one of the important operations to be performed for the development of image classifier. This method is helpful especially in the medical field where collection of large volumes of data becomes tedious. Data augmentation is an approach to improve the strength of the dataset by applying specified transformations on the existing dataset to generate new data. In this work, horizontal flipping, zoom of 0.2 and height and width shift of 0.2 were applied on the capsule endoscopy images.

2.5 Classification process

In this work, we used cross entropy loss function and adam optimizer. Cross entropy loss function also known as log loss relates the probabilities of the ground truth with that of the model prediction[28]. Adam is one of the most popular choices of optimizers used by deep learning researchers. Adam stands for adaptive estimation. This optimizer can greatly aid in the process of classification[29].Early stopping is used to find the convergence of the model while training. It will stop the training at the point at which there is no improvement in the performance metrics of the trained model. All the models were trained for 25 epochs and with 225 steps per each epoch based on the equation. The batch size was set to 16.

2.6 Evaluation metrics

Sensitivity, precision, accuracy, AUC and loss are the metrics used to evaluate the process of classification. These metrics are estimated through True Positive (TP), True Negatives (TN), False Positive (FP) and False Negative (FN) values. TP is the count of both the model prediction and ground truth are positive. FP is the count of ground truth is negative but model prediction is positive. TN is the count of both the model prediction and ground truth are negative and FN is the count of the ground truth is positive but prediction is negative.

Accuracy is defined as the ratio of the correct predictions to the total predictions. It can be represented by Eq. 1.

\(Accuracy= \frac{TP+TN}{TP+FP+FN+TN}\)(Eq. 1)

Precision is defined as the negatives predicted correctly to that of the total negatives. It is also called as specificity and it is represented by Eq. 2.

\(Specificity= \frac{TN}{TN+FP}\)(Eq. 2)

Sensitivity is defined as the ratio of correctly predicted positives to that of the total positive predictions. This is also known as recall and is represented by Eq. 3.

\(Sensitivity=\frac{TP}{TP+FN}\)(Eq. 3)

AUC refers to the area under the curve. In this case, the curve is the P-R curve (Precision-Recall). This curve is obtained as a plot between the true positive rate and false positive rate.

2.7 Web application design and development environment

The integrated development environment (IDE) cannot be accessed by the doctor for the diagnosis. Consequently, a tool that can assist a doctor in making a diagnosis is required. We have created a web application that can classify images from a capsule endoscopy into ulcers, polyps, and esophagitis automatically. Python's stream lit library was used to create this web application. The web application classifies the input capsule endoscopy images automatically, assisting the doctor in diagnosing the illness. Figure 3 represents the front-end design for the developed web application.

This entire process has been implemented using the Python language (version 3.6.7) and using the Keras module from Tensor flow (version 2.8.0) as backend and was run on Google Collab IDE with 8GB RAM and 12GB NVIDIA K80 GPU. The process of classification was performed using the keras module from tensor flow. The pretrained versions of the models were obtained from an application package of tensor flow keras and the additional layers were appended to these pretrained models to get the final model

The GastroEffNetV1 was trained on the capsule endoscopy images on google collab. The described image augmentation transformations were applied to improvise the dataset. Train test split ratio of 0.86 was used for the dataset. Table 2 shows the image distribution used for the work. The training and validation images were made into directories with a batch size of 16.

Table 2

Training and validation sets for the capsule endoscopy dataset
GRADE	TESTING	TRAINING
0-Normal	200	1300
1-ulcerative colitis	200	1300
2-polyps	200	1300
3-esophagitis	200	1300

The default learning rate (1e-3) was set for the adam optimizer. The training images were fit onto the model and the model was being trained for 25 epochs. Then the model was evaluated for both the training and validation image datasets. The performance plots for the different classification metrics were obtained. Finally, the frontend design for the web app was generated and was hosted. Tables 3 and 4 represents the comparative analysis of the performance metrics for the several CNN backbones trained on the capsule endoscopy images for training and validation respectively.

Table 3

Comparison of Evaluation metrics of various CNN architectures used for training images. (The bold values represent the best values)
CNN Backbone	Loss	Accuracy	AUC	Precision	Recall
VGG16	0.1958	0.9625	0.9966	0.9635	0.9608
VGG19	0.1849	0.9669	0.9982	0.9689	0.966
ResNet50	0.1143	0.9831	0.9993	0.9833	0.9825
EfficientNetB2	0.1389	0.9742	0.9986	0.9759	0.9733
ResNet101	0.1256	0.9798	0.9989	0.9811	0.9796
EfficientNetB6	0.1459	0.9796	0.999	0.9811	0.9787
EfficientNetB0	0.1272	0.9779	0.9985	0.9781	0.9775
EfficientNetB7	0.1499	0.9794	0.9984	0.9802	0.9787
EfficientNetV2B0	0.1038	0.9863	0.9994	0.9867	9862
EfficientNetV2B1	0.0808	0.9838	0.9992	0.984	0.9833
GastroEffNetV1 (proposed model)	0.0939	0.9873	0.9996	0.9873	0.9871
EfficientNetV2B3	0.1234	0.981	0.999	0.9815	0.9808

Table 4

Comparison of Evaluation metrics of various CNN architectures used for validation images. The bold values represent the best values.
CNN Backbone	Loss	Accuracy	AUC	Precision	Recall
VGG16	0.2089	0.955	0.9968	0.9585	0.9525
VGG19	0.1855	0.9625	0.9985	0.9637	0.9613
ResNet50	0.1276	0.98	0.9994	0.9812	0.98
EfficientNetB2	0.1709	0.965	0.9972	0.9662	0.9638
ResNet101	0.1281	0.9762	0.9993	0.9775	0.9762
EfficientNetB6	0.1472	0.9775	0.9993	0.9799	0.9762
EfficientNetB0	0.1366	0.975	0.9976	0.975	0.9737
EfficientNetB7	0.144	0.9862	0.9978	0.9887	0.985
EfficientNetV2B0	0.1047	0.9887	0.9997	0.9887	-
EfficientNetV2B1	0.0829	0.985	0.9987	0.985	0.985
GastroEffNet V1(Proposed model)	0.0918	0.9915	0.991	0.9925	0.9925
EfficientNetV2B3	0.1246	0.98	0.9979	0.98	-

GradCAM (gradient class activation mapping) technique has been used to derive the heat maps for the input image fed to the Convolutional Neural Network. The resultant map highlights the important regions at which the trained neural network focuses for classification. Figure 4 represents the heat maps for the different intestinal abnormalities obtained using GradCAM methodology. The heat map varies from blue to red, where blue means ‘lowest focus’ and red means ‘highest focus’.

The GastroEffNetV1 trained for 45 epochs using Adam optimizer and cross entropy loss function is chosen as the best model, since it performed the best in both training and validation dataset. Table 4 and Fig. 5 represent the performance plots for the proposed model (GastroEffNetV1).

Table 5

Evaluation metrics of the proposed model used for training and validation images.
METRIC	TRAINING	VALIDATION
Loss	0.0939	0.0918
Accuracy	0.9873	0.9915
Precision	0.9873	0.9925
Recall	0.9871	0.9925
AUC	0.9996	0.991

Figure 6 represents the input capsule endoscopy image and the corresponding prediction given by the model obtained from the website. The normal condition is considered as success and hence it is displayed in green color, whereas the rest are diseased conditions are considered as error and hence they are displayed in red color.

Li, B., & Meng, M. Q. H. developed a MLP neural network. Training MLP to detect abnormal regions in WCE images can be achieved by adjusting the weights in the neural network through gradient descent approach with respect to an error correction strategy. It is well known that MLP can be efficiently trained using backpropagation algorithm. This method produced a specificity of 87.81% [19].Tsuboi, A., et al developed a deep learning approach based on the Single Shot MultiBox Detector algorithm. This technique can extract specific features and quantities from a training data set, using multiple network layers (convolutional layers and pooling layers) and a back propagation algorithm. The outcome included the ability of CNN to correctly distinguish the location of angioectasia. The developed model achieved an AUC of 0.998 [20].

Alaskar, H., et al developed a pre-trained neural network GoogLeNet to recognize ulcer images by adding four new layers to the standard architecture, specifically, a dropout layer of 0.5 probability of dropout, a fully connected layer, a softmax layer and a classification-output layer. In the experiments, a total of 144 layers were used to build GoogLeNet. The developed model achieved an accuracy of 85% [21]. Hassan. A.R. et al developed a texture-feature-descriptor-based algorithm that operates on the Normalized Gray Level Co-occurrence Matrix (NGLCM) of the magnitude spectrum of the images to detect bleeding and non-bleeding areas. Normalized Gray Level Co-occurrence Matrix (NGLCM) matrix was constructed to extract features from each log transformed magnitude spectrum. They used linear SVM, polynomial and Radial Basis Function (RBF) kernels SVM classifiers. Among these linear SVM produced highest specificity of 98.95% [22].

Ghosh, T., et al developed computer-aided diagnostic (CAD) tools for automated analysis of detecting small intestinal abnormalities like bleeding. A VGG16 architecture is used as first 13 convolutional layers, a softmax classifier is fed by the decoder output feature maps for pixel-wise classification. the first 22 layers of our pre-trained AlexNet architecture are used, and the last three layers are designed according to our application. One fully connected layer is introduced that has two categories (bleeding and non-bleeding), thus contains two fully connected neurons. Also, one softmax and one classification output layer are included. This network is trained and tested on a single NVIDIA GeForce GTX 1070. In SegNet architecture, a VGG16 network is used as the first 13 convolutional layers and five max-pooling and five up-sampling layers are used. Finally, a soft-max classifier is fed by the decoder output feature maps for pixel-wise classification. This method produced a 98.49%F1 score and 97.51% sensitivity [23].

Wang, S., et al proposed a method called second glance (SecG) detection framework for automatic detection of ulcers using deep convolutional network. The architecture used was RetinaNet and also had a second glance refinement stage. The Refinement stage is built by the Resnet-34 and Resnet-50 backbone. With two CNN backbones and two anchor settings, we can have 4 different primary detectors RT34 and RT50 meaning RetinaNet with ResNet-34 and ResNet-50, respectively. A1 and A2 refer to anchor settings. This method produced a specificity of 90.48% [24]. Table 6 represents the comparison of validation accuracy and loss of several existing works with that of our work.

Table 6

Comparison of the validation accuracy of standard works with that of our work.
Work	Architecture Used	Accuracy
Li, B., & Meng, M. Q. H. [19]	MLP	88%
Wang, S[24]	CNN(ResNet-34)	90.1%
Ghosh, T., et al[23]	CNN (AlexNet and SegNet)	94.42%
Tsuboi, A., et al [20]	CNN	95%
Hassan. A.R[22]	SVM	99%
GastroEffNetV1 (Proposed model)	GastroEffNetV1	99.15%

Hence in this work we are able to develop a website that can detect gastrointestinal disorders like ulcers, esophagitis, polyps from capsule endoscopy images. This can act as a potent screening tool that aids physician to detect and diagnose such disorders. Since ulcers, esophagitis and polyps have high prevalence rates and development of such tool can be very useful. Using this tool, one can automatically detect the disease or disorder in the GI tract at low cost. Furthermore, with improved accuracy we can extract more information from the wireless capsule endoscopy images and can enhance the process. Some of the future works would be the combination with image enhancement algorithms and convolutional neural networks to improve upon the accuracy and to reduce the rate of misclassification in order to develop a better classification tool.

Ethical Approval

No human or animal is used in the study. Ethical clearance is not required.

Competing interests

All authors declare that no compete interest is in the study.

Authors' contributions

A, B, C, D & E wrote the main manuscript. B, C, D prepared figures 1-4, F. G prepared figures 5 & 6. A, B, D, F prepared the tables 1-6, B, C, E, F has done the literature survey, A, D & G analyzed the results and all the authors reviewed the manuscript.

Funding

No funding is associated with the study

Availability of data and materials

The data is open source data.

Conflict of Interest

All authors declare no conflict of interest. This work does not include funding from any external sources. All the authors were able to manage the expenses incurred throughout the research.

Fukudo, S., Okumura, T., Inamori, M., Okuyama, Y., Kanazawa, M., Kamiya, T., & Koike, K. (2021). Evidence-based clinical practice guidelines for irritable bowel syndrome 2020. Journal of gastroenterology, 56(3), 193-217. DOI:https://doi.org/10.1007/s00535-020-01746-z
Kobayashi, T., Siegmund, B., Le Berre, C., et al(2020). Ulcerative colitis. Nature reviews. Disease primers, 6(1), 74.DOI: https://doi.org/10.1038/s41572-020-0205-x
Baumgart, D. C., & Sandborn, W. J. (2012). Crohn's disease. The Lancet, 380(9853), 1590-1605.DOI:https://doi.org/10.1016/S0140-6736(12)60026-9
Du, L., & Ha, C. (2020). Epidemiology and Pathogenesis of Ulcerative Colitis. Gastroenterology clinics of North America, 49(4), 643–654. DOI:https://doi.org/10.1016/j.gtc.2020.07.005
Angarita, F. A., Feinberg, A. E., Feinberg, S. M., et al(2018). Management of complex polyps of the colon and rectum. International journal of colorectal disease, 33(2), 115-129.DOI:https://doi.org/10.1007/s00384-017-2950-1
Vitale, S. G., Haimovich, S., Laganà, et al(2021). Endometrial polyps. An evidence-based diagnosis and management guide. European journal of obstetrics, gynecology, and reproductive biology, 260, 70–77. DOI:https://doi.org/10.1016/j.ejogrb.2021.03.017
Antunes, C., & Sharma, A. (2021). Esophagitis. In StatPearls [Internet]. StatPearls Publishing.
Hiorns, M. P. (2011). Gastrointestinal tract imaging in children: current techniques. Pediatric Radiology, 41(1), 42-54.DOI: https://doi.org/10.1007/s00247-010-1743-2
Singh, R., Sathananthan, D., Tam, W., & Ruszkiewicz, A. (2013). Endocytoscopy for diagnosis of gastrointestinal Neoplasia: the expert's approach. Video Journal andEncyclopediaofGIEndoscopy,1(1),18-19.DOI:https://doi.org/10.1016/S2212-0971(13)70009-8
Mensink, P. B. F., Haringsma, J., Kucharzik, T., et al(2007). Complications of double balloon enteroscopy: a multicenter survey. Endoscopy, 39(07), 613-615.DOI: 10.1055/s-2007-966444
Nakamura, T., &Terano, A. (2008). Capsule endoscopy: past, present, and future. Journal of gastroenterology, 43(2), 93-99. DOI: 10.1007/s00535-007-2153-6
Wang, S., Xing, Y., Zhang, L., et al (2019). A systematic evaluation and optimization of automatic detection of ulcers in wireless capsule endoscopy on a large dataset using deep convolutional neural networks. Physics in Medicine & Biology, 64(23), 235014.DOI: 10.1088/1361-6560/ab5086
Barash, Y., Azaria, L., Soffer, S., et al(2021). Ulcer severity grading in video capsule images of patients with Crohn’s disease: An ordinal neural network solution. Gastrointestinal Endoscopy, 93(1), 187-192. DOI: 10.1016/j.gie.2020.05.066
Saito, H., Aoki, T., Aoyama, K., et al (2020). Automatic detection and classification of protruding lesions in wireless capsule endoscopy images based on a deep convolutional neural network. Gastrointestinal endoscopy, 92(1), 144-151.DOI: 10.1016/j.gie.2020.01.054
Ghosh, T., &Chakareski, J. (2021). Deep transfer learning for automated intestinal bleeding detection in capsule endoscopy imaging. Journal of Digital Imaging, 34(2), 404-417.DOI:10.1007/s10278-021-00428-3Yuan, Y., Wang, J., Li, B., & Meng, M. Q. H. (2015).
Saliency based ulcer detection for wireless capsule endoscopy diagnosis. IEEE transactions on medical imaging, 34(10), 2046-2057.DOI:10.1109/TMI.2015.2418534
Pogorelov, K., Randel, K. R., Griwodz, C., et al (2017, June). Kvasir: A multi-class image dataset for computer aided gastrointestinal disease detection. In Proceedings of the 8th ACM on Multimedia Systems Conference (pp. 164-169).DOI:https://doi.org/10.1145/3083187.3083212
Silva, J., Histace, A., Romain, O.,et al (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), 283-293. DOI: 10.1007/s11548-013-0926-3
Li, B., & Meng, M. Q. H. (2009). Computer-based detection of bleeding and ulcer in wireless capsule endoscopy images by chromaticity moments. Computers in biology and medicine, 39(2), 141-147. DOI: 10.1016/j.compbiomed.2008.11.007
Tsuboi, A., Oka, S., Aoyama, K., et al(2020). Artificial intelligence using a convolutional neural network for automatic detection of small‐bowel angioectasia in capsule endoscopy images. Digestive Endoscopy, 32(3), 382-390.DOI: 10.1111/den.13507
Alaskar, H., Hussain, A., Al-Aseem, N., et al(2019). Application of convolutional neural networks for automated ulcer detection in wireless capsule endoscopy images. Sensors, 19(6), 1265. DOI:https://doi.org/10.3390/s19061265
Hassan, A. R., & Haque, M. A. (2015). Computer-aided gastrointestinal hemorrhage detection in wireless capsule endoscopy videos. Computer methods and programs in biomedicine, 122(3), 341-353.DOI: 10.1016/j.cmpb.2015.09.005
Ghosh, T., &Chakareski, J. (2021). Deep transfer learning for automated intestinal bleeding detection in capsule endoscopy imaging. Journal of Digital Imaging, 34(2), 404-417.DOI: 10.1007/s10278-021-00428-3
Wang, S., Xing, Y., Zhang, L., Gao, H., & Zhang, H. (2019). Deep convolutional neural network for ulcer recognition in wireless capsule endoscopy: experimental feasibility and optimization. Computational and mathematical methods in medicine, 2019.DOI:https://doi.org/10.1155/2019/7546215
Lu, M., Zhao, Q., Zhang, J., Pohl, K. M., Fei-Fei, L., Niebles, J. C., & Adeli, E. (2021). Metadata Normalization. Proceedings. IEEE Computer Society Conference on Computer Vision and Pattern Recognition, 2021, 10912–10922. DOI:https://doi.org/10.1109/cvpr46437.2021.01077
Khan, S. H., Hayat, M., &Porikli, F. (2019). Regularization of deep neural networks with spectral dropout. Neural networks :the official journal of the International Neural Network Society, 110, 82–90.DOI:https://doi.org/10.1016/j.neunet.2018.09.009
Tan, M., & Le, Q. (2019, May). Efficientnet: Rethinking model scaling for convolutional neural networks. In International conference on machine learning (pp. 6105-6114). PMLR. .DOI:https://doi.org/10.48550/arXiv.1905.11946
Rajaraman, S., Zamzmi, G., &Antani, S. K. (2021). Novel loss functions for ensemble-based medical image classification. Plos one, 16(12), e0261307.DOI: https://doi.org/10.1371/journal. pone.0261307
Bock, S., Goppold, J., & Weiß, M. (2018). An improvement of the convergence proof of the ADAM-Optimizer. arXiv preprint arXiv:1804.10587. DOI:https://doi.org/10.48550/arXiv.1804.10587
Russo, F. (2003). A method for estimation and filtering of Gaussian noise in images. IEEE Transactions on Instrumentation and Measurement, 52(4), 1148-1154. DOI:https://doi.org/10.1109/TIM.2003.815989.

No competing interests reported.

Download PDF

Version 1

posted

You are reading this latest preprint version

GastroEffNetV1- CNN based Automated detection of Gastrointestinal abnormalities from capsule endoscopy images

Status:

Version 1

Abstract

Figures

1. Introduction:

1.1 Related Work

2. Materials And Methods

2.1 Material Description

2.2 Designated Workflow

2.3 Architecture for classifier model

2.4 Data Augmentation

2.5 Classification process

2.6 Evaluation metrics

2.7 Web application design and development environment

3 Results

4 Discussions

5. Conclusion

Declarations

References

Additional Declarations

Status:

Version 1