Modified fuzzy based neuro networks for the prediction of common thorax diseases

• Published in: Multimedia tools and applications Vol. 83; no. 40; pp. 87479 - 87503
• Main Authors: Kumar, C. Ashok, Priya, R. Lakshmi, Ambika, I., Mahiba, C.
• Format: Journal Article
• Language: English
• Published: New York Springer US 01.12.2024; Springer Nature B.V
• Subjects: Accuracy; Algorithms; Artificial intelligence; Chest; Clustering; Computer Communication Networks; Computer Science; Data Structures and Information Theory; Diaphragm (anatomy); Diaphragm wall; Disease; Disorders; Esophagus; Fuzzy logic; Mediastinum; Modelling; Multimedia Information Systems; Neural networks; Prediction models; Predictions; Special Purpose and Application-Based Systems; System identification; Thorax; Track 2: Medical Applications of Multimedia; DenseNet; De-fuzzification; Fuzzification; Normalization
• ISSN: 1573-7721; 1380-7501; 1573-7721
• DOI: 10.1007/s11042-024-18831-7

In the realm of artificial intelligence, fuzzy logic emerges as a valuable tool for predicting thoracic disorders, encompassing various medical conditions affecting the heart, lungs, mediastinum, esophagus, chest wall, major vessels, and diaphragm. This predictive system relies on an objective fuzzy modeling approach, which has proven highly effective in enhancing accuracy by integrating clustering algorithms with fuzzy system identification. To train this predictive engine, historical data pertaining to common thoracic disorders is gathered from reliable online sources. Relevant data are meticulously collected and processed, retaining only essential inputs for the prediction system. The recorded data undergoes logical processing and is normalized through a fuzzification process. The prediction model leverages the power of DenseNet, particularly in "reading chest X-rays," enabling the identification and localization of prevalent disease patterns using image-level labels alone. The performance of the Modified Fuzzy-Based Neural Networks for predicting common thoracic diseases is remarkable, achieving an impressive accuracy rate of 96% with just 10 training epochs and with training and validation loss consistently below 5%. This innovative approach seamlessly combines objective fuzzy modeling with DenseNet utilization for thoracic disease prediction. By integrating historical data, meticulous preprocessing, normalization, disease prediction, and de-fuzzification, the system excels in recognizing and precisely locating frequently encountered thoracic diseases, offering a high degree of accuracy in diagnosis and assessment.