Exploiting probability density function of deep convolutional autoencoders' latent space for reliable COVID-19...

Area of Science:

• Medical Imaging Analysis
• Artificial Intelligence in Healthcare
• Infectious Disease Diagnostics

Background:

• Accurate and rapid diagnosis of COVID-19 is crucial for patient management and public health.
• Chest computed tomography (CT) scans are valuable for COVID-19 diagnosis, but automated classification methods are needed.
• Existing methods may require large labeled datasets or supervised training, posing challenges.

Purpose of the Study:

• To develop and evaluate an unsupervised probabilistic method for classifying chest CT scans as COVID-19 or non-COVID-19.
• To leverage deep convolutional autoencoders (DCAEs) for feature extraction and probability density estimation.
• To assess the performance of the proposed method against state-of-the-art techniques.

Main Methods:

• An unsupervised deep convolutional autoencoder (DCAE) was trained exclusively on COVID-19 CT scans.
• Kernel density estimation (KDE) was used to build a probability density function (PDF) from the hidden feature vectors.
• Test CT scans were encoded, their PDF values computed, and compared to a threshold for classification.

Main Results:

• The DCAE successfully generated compact hidden representations of COVID-19 CT scans.
• The probabilistic method, utilizing the learned PDF, demonstrated effective classification of test CT scans.
• Performance metrics, including test accuracy and training times, were numerically evaluated and compared to existing methods.

Conclusions:

• The proposed unsupervised probabilistic method offers a viable approach for COVID-19 classification using chest CT scans.
• This method demonstrates the potential of unsupervised learning with DCAEs and KDE for medical image analysis.
• The findings suggest a promising direction for developing automated diagnostic tools for infectious diseases.