Abstract

Diabetes Mellitus, known as Diabetes, is a prevalent chronic condition requires monitoring the
health condition of a Diabetic patient. Early identification of Diabetes Mellitus can help reduce
the risk of an individual to develop devastating complications and can delay or prevent the
Diabetes Mellitus condition. To avoid organ related health issues, a Diabetic patient monitors
blood glucose level and checks visible abnormalities on various human organs such eye, tongue,
mouth, foot, etc. But to check blood glucose level, a patient needs to take a blood sample for
the conventional medical test that is unpleasant, costly and challenging for an elderly individual
who may need help from another person. The portable smart device technology could enhance
the experience for Diabetic patients to screen and check health condition in their convenience.
Oral Candidiasis, Fissured, Glossitis and Geographic abnormality appear on tongue or tongue
surface because of uncontrolled or controlled blood glucose level in a human body and these
tongue abnormalities may appear during early onset of Diabetes Mellitus. A person with Dia�betes Mellitus condition is more likely to develop one or more abnormalities due to high glucose
level present in the blood. The medical test to measure blood glucose level is inconvenient for
patients living in remote areas where medical services are minimal or not available.
This thesis introduces ‘Diatonash’ database, a new Diabetic tongue database comprises Dia�betes Mellitus related clinical test results, digital images of four tongue abnormalities, various
signs and symptoms, and clinical diagnosis outcomes of secondary diseases complied with west�ern medicine in practice. A streamlined, coherent and efficient clinical diagnosis procedures
have been developed and introduced to screen and diagnose the Diabetic tongue abnormalities
complying with western medicine in practice. The new clinical diagnosis procedures (also known
as systemic clinical examination procedures) consider two categories of tongue features such as
distinctive features and shared features of the Diabetic tongue abnormalities. Then, the visual
features on tongue extracted manually from 572 digital tongue images of 166 patients following
the new coherent diagnosis procedures.
To screen and check human tongue condition, the proposed computer-aided diagnosis model
classifies and predicts abnormalities appear on tongue or tongue surface considering visual signs
cause by Diabetes Mellitus. The Naïve Bayes and Random Forest classifier applied on manually
extracted visible tongue features to train and construct the computer-aided diagnosis models.

Four single models for the four Diabetic tongue abnormalities and one single model for the
same four abnormalities were developed to classify and predict Diabetic tongue abnormalities.
The accuracy of the four single machine learning models range from 95.8% to 100% and accuracy
of the single machine learning model is 97%. One of the key findings is Random Forest algorithm
performed better than Naïve Bayes algorithm to classify four Diabetic tongue abnormalities
simultaneously. The proposed computer-aided diagnosis model is convenient to screen and check
Diabetes Mellitus related visible health issues reducing cost for the medical test and accessible
to patients residing in the remote areas.