Fuzzy and Rule-based Hybrid Expert System for Accurate Diagnosis of Tuberculosis

Abstract

Starting from many decades back till recent years, there have been diseases which affect and cause numerous deaths in the human race. Among the deadliest diseases known to man, tuberculosis (TB) is one with a very high mortality rate. One third of the world’s population is affected by the disease. TB is one of the top ten deadliest diseases in Ethiopia having the fourth place next to HIV. Poverty, malnutrition, over-crowded living conditions and high prevalence of HIV infection, which are indicators of low and middle income countries like Ethiopia, are some of the risk factors which increase the transmission level. In the current manual diagnosis of TB, misdiagnosis and diagnostic delays are being witnessed. This is mainly due to TB having different classifications, its nature of mimicking other diseases and the disproportionate number of human TB experts and patients. Even though there are expert systems designed in foreign countries, their questionable accuracy, high technology as well as knowledge requirements and resourceful settings makes them unfit and inapplicable for developing countries. The objective of the current thesis is to design a diagnostic decision support system for accurate diagnosis of tuberculosis by using fuzzy and rule based hybrid expert systems. Patient diagnosis data was collected from eighty patients, from which seventy five percent of the data was used to formulate the rules used in the system. By using the fuzzy expert system, the patients’ symptom severity range was analysed to indicate the suspicion level of each disease. In the fuzzy expert system, fuzzification of the crisp input was made by using triangular membership function and mamdani fuzzy inference mechanism was applied to map the given input variable to an output space. The center of gravity defuzzification method was used in the inference process to obtain a crisp final output for each disease’s level of suspicion. Once the disease with the highest suspicion level was found, different examinations were conducted to confirm the suspicion and the findings were analysed by using rule based expert system. In the rule based expert system, by using TB experts’ knowledge, rules were formulated and forward chaining inference strategy was applied to reach at a diagnostic conclusion. Among the examination modalities that are used to diagnose pulmonary TB, the chest X-ray classification is made by the system. To design the classification model, two chest X-ray image datasets were used. Data augmentation was applied on the training data to increase the number and variation of data. The model was trained with ResNet50, a pre-trained convolutional network using 80% of the data. Fuzzy and Rule based hybrid Expert System for accurate diagnosis of Tuberculosis /2020 III The model was found to have 84% classification accuracy. An interactive user interface was also designed by using visual studio which will make the system more user-friendly. A computerized patient data recording system which will facilitate TB patient follow-up, keep their data safe and make it easily accessible was also incorporated. The database is designed using SQL server. Finally, after conducting a performance evaluation on the designed fuzzy and rule based hybrid expert system, the system achieved an accuracy of 93%. This result suggests the success of the expert system in increasing the accuracy of tuberculosis disease diagnosis in the absence of human TB experts.