TMV-NET: Regression approach for the quantification of Tympanic Membrane

Tympanic membrane, an essential part of the human ear whose function is to help in hearing. In this paper, novelty-based extraction of tympanic membrane blood vessels using TMV-Net is performed. The neural network layers are built in a manner where most features are retained at every convolution block. The prediction of the bounding box values is more accurate as the resolution of the image is retained at every convolution layer. The image is further upscaled with the help of SR-GAN neural networks. The output image is sharpened and hence used for quantifying the blood vessels with the help of histogram approach. The histogram approach considers the use of green channel pixels over the total area of the image giving the percentage of blood vasculature.