Automatic Detection of Early Esophageal Cancer from Endoscope Image Using Fractal Dimension and Discrete Wavelet Transform

Abstract

We propose a new method for automatically detecting early esophageal cancer from an endoscopic image. We decompose the original image into four components, namely, the RGB and luminance components, and apply the discrete wavelet transform (DWT) to these components twice. The fractal dimensions are computed at each small block using the box-counting method, and the abnormal regions are detected based on the fractal dimensions. In addition, to process the endoscopic image quickly, we clip the portion that does not contain the esophageal cancer from the original endoscopic image and resize the remaining image to 1024 × 1024 pixels by mirroring. Finally, we show the abnormal region by using the product of the computed fractal dimensions from the four components. Our method is not intended for providing accurate diagnosis of the detected abnormal regions as esophageal cancer, but is intended to provide additional information to help doctors in their diagnosis. Therefore, our method only needs to detect all of the regions suspected of being cancer even if the detect results are false positives. We describe the procedure used in our method in detail and present experimental results demonstrating that our method is able to detect abnormal regions suspected of being early esophageal cancer using practical endoscopic images.

Automatic esophagus Z‐line delineation in endoscopic images using a new boundary linking method

“…However, BE and EAC regions are more recognizable than the regions that suffer from early-stage Barrett (our case) due to their tissue deformation and/or colour changes. Therefore, the methods presented in [28][29][30][31][33][34][35][36] could not help annotate the Z-line in our dataset. Along with this reason, the deep learning-based methods provided in [13,18,32] cannot be suitable due to the difference in the quality of the images.…”

“…The segmentation method is applied to the images with remarkable tissue changes. Yamaguchi et al [35] proposed a method based on the discrete wavelet transform and fractal dimension to diagnose the suspected area. The images used in this manuscript are taken using the new laser endoscope system and are useful for detecting the early cancerous area with colour or texture changes.…”

“…According to the results, the dice, recall, and precision values are 0.77, 0.75, and 0.77, respectively. Any evaluation method is used in [35] and [36] to investigate the performance of the proposed annotation methods. The suggested method for Barrett's mucosa segmentation in [37] is applied to four images as the dataset, and the value of the average accuracy for these images is calculated as 0.95.…”

High Accuracy Thyroid Tumor Image Recognition Based on Hybrid Multiple Models Optimization

“…Texture features include ultrasound appearance and echo model. Traditional feature extraction mainly includes shape [12], color [13] and texture [12], [14]. In addition, SIFT and HOG [15] are also commonly used in feature extraction of image recognition.…”