Greedy Autoaugment for classification of mycobacterium tuberculosis image via generalized deep CNN using mixed...

Area of Science:

• Medical Imaging
• Computational Biology
• Infectious Diseases

Background:

• Tuberculosis (TB) remains a global health challenge, with current diagnostic methods failing to identify all infected individuals, leading to continued transmission.
• Accurate and early diagnosis is crucial for effective TB control and preventing societal spread.

Purpose of the Study:

• To develop an accurate image-based processing system for the early screening and diagnosis of tuberculosis (TB).
• To classify Mycobacterium tuberculosis in microscopic images using an improved and generalized convolutional neural network (CNN).

Main Methods:

• A dataset of 1078 negative and 469 positive Mycobacterium tuberculosis images was utilized.
• Preprocessing involved Square Rough Entropy (SRE) thresholding to remove irrelevant image parts.
• Greedy AutoAugment selected top data augmentation policies to prevent overfitting, and mixed pooling enhanced CNN generalization.

Main Results:

• The proposed generalized CNN achieved a 93.4% accuracy in classifying Mycobacterium tuberculosis images.
• Techniques like generalized pooling, batch normalization, Dropout, and PReLU significantly improved classification performance.
• The CNN outperformed other classifiers, including Naïve Bayes-LBP, KNN-LBP, GBT-LBP, Naïve Bayes-HOG, KNN-HOG, SVM-HOG, and GBT-HOG.

Conclusions:

• The improved CNN model demonstrates significant potential for the accurate diagnosis of tuberculosis from microscopic images.
• This AI-driven approach offers a promising tool to enhance early TB detection and combat disease transmission.