Dental Imaging Project – :/P (Tongue Segmentation)

Hi ! This is Me (Meghasyam),  from the Dental Imaging Project being mentored by Dr. Hyunsung Park (Awesome Guy !). We at the Dental Imaging project have been working on separating tongue from images of the tongue. Various clinical studies have shown correlation between the abnormalities of tongue(its coating, structure) and, visceral cancers and heart diseases[1]. The segmented tongue can also help in estimating Hemoglobin levels in the body, and also any deformation/disease on the tongue.

We assumed that we have an image of the tongue that is oriented in such a way that the tongue is present in the bottom part of the image. We have implemented two algorithms C2G2FSnakes[2] and DGF(Double Geo-Vector Flow)[1]. Basically what both the algorithms do is –

1. Initialize the contour based on edge detection

2. Iteratively converge to the real boundary by minimizing the energy function[1][2].

One problem that arose when iterating the contour is that some points on it go out of the bounds of the image. This was corrected by reinitializing them to the values of the previous iteration.

In the world of selfies with people putting their tongues out, we decided to test out one(:P un intended). Below is the photo of the tongue selfie of one of my teammates.


The Tongue Selfie

The results of the algorithms applied on the image cropped from the selfie are as follows –


Tongue Segmentation using C2G2FSnake algorithm


Tongue Segmentation using DGFSnake

[1] Computerized tongue image segmentation via the double geo-vector flow, Miao-Jing Shi, Guo-Zheng Li, Fu-Feng Li and Chao Xi, Chinese Medicine.

[2] C2G2FSnake: Automatic tongue image segmentation using prior knowledge, Miao-Jing Shi, Guo-Zheng Li, Fu-Feng, Science China Information Sciences.


The Team: Hyunsung Park (mentor), Avi Singh, Ujjwal Baid, Raghu, Meghasyam, Ashish Bora, Ashray Malhotra, Pratima Mokashi, Samriddh Hada