Semi-Supervised Segmentation Framework for Gastrointestinal Lesion Diagnosis in Endoscopic Images

BACKGROUND

Accurate gastrointestinal (GI) lesion segmentation is crucial in diagnosing digestive tract diseases. An automatic lesion segmentation in endoscopic images is vital to relieving physicians' burden and improving the survival rate of patients. However, pixel-wise annotations are highly intensive, especially in clinical settings, while numerous unlabeled image datasets could be available, although the significant results of deep learning approaches in several tasks heavily depend on large labeled datasets. Limited labeled data also hinder trained models' generalizability under fully supervised learning for computer-aided diagnosis (CAD) systems.

METHODS

This work proposes a generative adversarial learning-based semi-supervised segmentation framework for GI lesion diagnosis in endoscopic images to tackle the challenge of limited annotations. The proposed approach leverages limited annotated and large unlabeled datasets in the training networks. We extensively tested the proposed method on 4880 endoscopic images.

RESULTS

Compared with current related works, the proposed method validates better results (Dice similarity coefficient = 89.42 ± 3.92, Intersection over union = 80.04 ± 5.75, Precision = 91.72 ± 4.05, Recall = 90.11 ± 5.64, and Hausdorff distance = 23.28 ± 14.36) on the challenging multi-sited datasets, confirming the effectiveness of the proposed framework.

CONCLUSION

We explore a semi-supervised lesion segmentation method to employ the full use of multiple unlabeled endoscopic images to improve lesion segmentation accuracy. Experimental results confirmed the potential of our method and outperformed promising results compared with the current related works. The proposed CAD system can minimize diagnostic errors.

Annotating Early Esophageal Cancers Based on Two Saliency Levels of Gastroscopic Images

Early diagnoses of esophageal cancer can greatly improve the survival rate of patients. At present, the lesion annotation of early esophageal cancers (EEC) in gastroscopic images is generally performed by medical personnel in a clinic. To reduce the effect of subjectivity and fatigue in manual annotation, computer-aided annotation is required. However, automated annotation of EEC lesions using images is a challenging task owing to the fine-grained variability in the appearance of EEC lesions. This study modifies the traditional EEC annotation framework and utilizes visual salient information to develop a two saliency levels-based lesion annotation (TSL-BLA) for EEC annotations on gastroscopic images. Unlike existing methods, the proposed framework has a strong ability of constraining false positive outputs. What is more, TSL-BLA is also placed an additional emphasis on the annotation of small EEC lesions. A total of 871 gastroscopic images from 231 patients were used to validate TSL-BLA. 365 of those images contain 434 EEC lesions and 506 images do not contain any lesions. 101 small lesion regions are extracted from the 434 lesions to further validate the performance of TSL-BLA. The experimental results show that the mean detection rate and Dice similarity coefficients of TSL-BLA were 97.24 and 75.15%, respectively. Compared with other state-of-the-art methods, TSL-BLA shows better performance. Moreover, it shows strong superiority when annotating small EEC lesions. It also produces fewer false positive outputs and has a fast running speed. Therefore, The proposed method has good application prospects in aiding clinical EEC diagnoses.

Computer assisted gastric abnormalities detection using hybrid texture descriptors for chromoendoscopy images

BACKGROUND AND OBJECTIVE

The early diagnosis of stomach cancer can be performed by using a proper screening procedure. Chromoendoscopy (CH) is an image-enhanced video endoscopy technique, which is used for inspection of the gastrointestinal-tract by spraying dyes to highlight the gastric mucosal structures. An endoscopy session can end up with generating a large number of video frames. Therefore, inspection of every individual endoscopic-frame is an exhaustive task for the medical experts. In contrast with manual inspection, the automated analysis of gastroenterology images using computer vision based techniques can provide assistance to endoscopist, by finding out abnormal frames from the whole endoscopic sequence.

METHODS

In this paper, we have presented a new feature extraction method named as Gabor-based gray-level co-occurrence matrix (G2LCM) for computer-aided detection of CH abnormal frames. It is a hybrid texture extraction approach which extracts a combination both local and global texture descriptors. Moreover, texture information of a CH image is represented by computing the gray level co-occurrence matrix of Gabor filters responses. Furthermore, the second-order statistics of these co-occurrence matrices are computed to represent images' texture.

RESULTS

The obtained results show the possibility to correctly classifying abnormal from normal frames, with sensitivity, specificity, accuracy, and area under the curve as 91%, 82%, 87% and 0.91 respectively, by using a support vector machine classifier and G2LCM texture features.

CONCLUSION

It is apparent from results that the proposed system can be used for providing aid to the gastroenterologist in the screening of the gastric tract. Ultimately, the time taken by an endoscopic procedure will be sufficiently reduced.

Identification of lesion images from gastrointestinal endoscope based on feature extraction of combinational methods with and without learning process. Med Image Anal. 2016;32:281-294. [PMID: 27236223 DOI: 10.1016/j.media.2016.04.007]

The gastrointestinal endoscopy in this study refers to conventional gastroscopy and wireless capsule endoscopy (WCE). Both of these techniques produce a large number of images in each diagnosis. The lesion detection done by hand from the images above is time consuming and inaccurate. This study designed a new computer-aided method to detect lesion images. We initially designed an algorithm named joint diagonalisation principal component analysis (JDPCA), in which there are no approximation, iteration or inverting procedures. Thus, JDPCA has a low computational complexity and is suitable for dimension reduction of the gastrointestinal endoscopic images. Then, a novel image feature extraction method was established through combining the algorithm of machine learning based on JDPCA and conventional feature extraction algorithm without learning. Finally, a new computer-aided method is proposed to identify the gastrointestinal endoscopic images containing lesions. The clinical data of gastroscopic images and WCE images containing the lesions of early upper digestive tract cancer and small intestinal bleeding, which consist of 1330 images from 291 patients totally, were used to confirm the validation of the proposed method. The experimental results shows that, for the detection of early oesophageal cancer images, early gastric cancer images and small intestinal bleeding images, the mean values of accuracy of the proposed method were 90.75%, 90.75% and 94.34%, with the standard deviations (SDs) of 0.0426, 0.0334 and 0.0235, respectively. The areas under the curves (AUCs) were 0.9471, 0.9532 and 0.9776, with the SDs of 0.0296, 0.0285 and 0.0172, respectively. Compared with the traditional related methods, our method showed a better performance. It may therefore provide worthwhile guidance for improving the efficiency and accuracy of gastrointestinal disease diagnosis and is a good prospect for clinical application.