Development of a hybridized artificial neural network, fuzzy logic and particle swarm optimization model for improved diagnosis of tuberculosis

Abstract

Background

Tuberculosis (TB), a highly infectious disease caused by Mycobacterium tuberculosis, remains a leading cause of mortality, particularly in resource-limited regions. Limitations in sensitivity, specificity, and diagnostic delays often challenge traditional diagnostic methods. To address these, we propose a hybrid diagnostic model based on Particle Swarm Optimization- Aided Neuro-Fuzzy Model (PSOFUZ) which integrates fuzzy logic, artificial neural networks (ANN), and particle swarm optimization (PSO).

Materials and methods

The dataset used for the study was obtained from the Center for Disease Control and Prevention. The dataset contains 1200 records, of which 850 were TB-positive and 350 were TB-negative. The dataset included relevant clinical and demographic attributes such as smoking status, drug use, HIV status, body temperature, blood pressure, and prior history of tuberculosis. These inputs were fuzzified with Gaussian membership functions, which transform clinical data into fuzzy values that capture uncertainty and imprecision. The fuzzified data was then fed into an ANN, which analyzes patterns and detects potential TB symptoms. PSO was used to optimize both the ANN's connection weights and the parameters of the fuzzy membership functions

Result

The model gave an accuracy of 91.2 %, sensitivity of 93.4%, and specificity of 89.1%, outperforming conventional and other computational models (Neuro-Fuzzy: 88.9%, 83.5% and 86.2% of sensitivity, specificity and accuracy respectively). Our results hence show that PSOFUZ outperformed the baseline models in terms of accuracy, sensitivity, and specificity, as the experimental data show. When compared to non-optimized or single-method techniques, these improvements show that the joint optimization of neural and fuzzy parameters adds to improved generalization and more reliable tuberculosis diagnosis.

Conclusion

This study demonstrated the potential of intelligent hybrid models to enhance TB diagnosis, especially in settings where systems blend fuzzy logic interpretability with neural network adaptive learning capabilities, resulting in increased diagnostic accuracy and resilience with diagnostic resource constraints.