Dual-Path Mobile Net-Based Framework for Esophageal Cancer Detection using Enhanced Endoscopic Imaging

Esophageal Cancer (EC) is essential for increasing patient survival in early identification; yet, endoscopists find it difficult to identify the cancer cells. This paper proposed a Dual-Path MobileNet for esophageal cancer screening is proposed to reduce the workload of physician and increase detection accuracy. Firstly, un-sharp mask filter is applied to Esophageal Endoscopy Images to brighten the edge
and to increase its sharpness for better quality of image. Next, the processed image is given to segmentation process using Self Organizing Map (SOM) clustering algorithm. Here the SOM convert the high-resolution image into a lower-resolution image. After that image is extracted using Histogram of Oriented Gradient (HOG) it capture the edge and shape of the gradient direction for further analysis.
Finally, a Dual-path MobileNet framework is processed to enhance the classification of esophageal cancer diagnosis. Using python software the proposed framework have an improved accuracy of 95% is accomplished when compared to other techniques.