Automatic PAP Mask Sizing with an Error Correcting Autoencoder.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Machine Learning

Background:

• Accurate sizing of Positive Airway Pressure (PAP) masks is crucial for effective sleep apnea treatment.
• Current image-based automatic sizing methods face challenges with accuracy and overfitting, especially with limited custom datasets.

Purpose of the Study:

• To improve the accuracy of image-based automatic PAP mask sizing.
• To mitigate overfitting in deep learning models used for mask sizing.

Main Methods:

• A convolutional neural network (CNN) was pre-trained on the MUCT dataset using transfer learning.
• An error correcting autoencoder was integrated into the CNN architecture.
• The augmented model was trained on a custom dataset for PAP mask sizing.

Main Results:

• The enhanced CNN model achieved a 15.3% increase in PAP mask sizing accuracy compared to the baseline model.
• The integration of the autoencoder significantly reduced overfitting.
• Improved overall performance in automatic PAP mask sizing.

Conclusions:

• The proposed error correcting autoencoder-augmented CNN model offers a significant advancement in automatic PAP mask sizing accuracy.
• This approach effectively addresses overfitting, leading to more reliable sizing predictions.
• The method holds promise for enhancing patient comfort and treatment efficacy in sleep apnea therapy.