Dental Imaging Project: Beyond the Teeth(Gums)
This is Sri Raghu from the Dental Imaging Project reporting about the Gingivitis and reverse smoker’s palate diseases detection algorithm we made as a part of our project. The first task that we (Me and Meghasyam) were assigned in our task is to detect gingivitis from the intra oral images.
What is Gingivitis?
It is a common and mild gum disease in which we can observe swollen reddish brown gums and causes mild irrtation the patients. The serious concern with this disease is that it being mild, you may not be aware of the existence of this disease and this might lead to other severe gum diseases and tooth loss. So, our project deals with the detection of gingivitis in its early stages and give a caution to the person.
Our method for detection:
We tried to solve this problem on intensity based approach. Some of the challenges we faced is that, we don’t have any standard dataset, to test our algorithms. And none of the team members are having this disease to prepare our own dataset ( 😛 ) . So we designed our algo on the basis of google images. The image shown above is a google image for the gingivitis disease.
To get the best inliers we need to remove maximum outliers. The first set of outliers in our image is the teeth, which if we succeed half of our starting problem for any tooth or gum detection is solved. For that we moved from the traditional RGB-Color space HSI-space. The above image is the HSI image of the above Gingivitis image. If we clearly observe, we can observe that the teeth are having a peculiar color when compared to the rest of the image. So we experimented with the values of the hue space and
intensity space of several images and got to the best threshold values to that will seperate not only the teeth but also the some part of the healthy gum. Our algorithm can work perfectly on removing the teeth and also some part of healthy gums but does not remove a single diseased gum. You can see the result in the image below. We tested our algo on 26 images we have in our dataset and it worked! Right now our algo can only segment the gum with unusual characteristics but cannot judge the probability of gingivitis. This is because we have to train our algo with a dataset before judging.
The last result is the fun part when we tested on Hyunsung’s teeth.
The Team: Hyunsung Park (mentor), Avi Singh, Ujjwal Baid, Raghu, Meghasyam, Ashish Bora, Ashray Malhotra, Pratima Mokashi, Samriddh Hada