The authors describe a method to process the information gathered in wireless capsule endoscopy (WCE) images and ultimately determine whether a region of interest (ROI) is benign, malignant, or normal.

The patient swallows the WCE device, which has a complementary metal–oxide–semiconductor (CMOS) sensor that can be used to gather digital images as it passes through the patient’s digestive system. The authors estimate the device can gather a total of 50000 digital images. Human analysis of these images can be tedious and time-consuming. The authors describe a procedure where these WCE images can undergo machine processing.

Each image undergoes the following steps:

(1) Image enhancement: since the WCE device is moving through the digestive system, noise can be introduced into an image as a result of motion. A Wiener filter is used to enhance the image and reduce the noise created by the motion. The ROI is then converted into RGB color by a color histogram. The output from these steps is then passed into an image clustering procedure.

(2) Image clustering via a k-means clustering method. This determines whether a ROI may be cancerous. The ROI undergoes feature extraction using a spatial gray level dependence method (SGLDM). In effect, a texture analysis of the ROI allows feature extraction. The output from this step is then passed to a tumor classification step.

(3) The tumor classification step uses two support vector machine classifiers (SVM1 and SVM2). SVM1 is used to determine whether the WCE image is normal or abnormal. Following this, the selected feature is analyzed by the SVM2 classifier to determine whether it is benign, malignant, or normal.

The use of WCE images does not eliminate the need for patient preparation. It is implied that the device has to pass through the patient, and if a tumor (benign or malignant) is detected, another medical procedure will need to be performed in order to analyze the tumor or polyp.

The authors compare their procedure to other methods used to evaluate WCE images and show that their method has greater accuracy. However, they do not compare their method to a standard colonoscopy, which is estimated to have the highest level of accuracy [1].