A Fully Automated System for Sizing Nasal PAP Masks Using Facial Photographs

A Method for Identifying Ground Truth Labels in Regression Problems using Annotator Precision

We propose a novel method for deriving ground truth labels for regression problems that considers the precision of annotators separately for each label. This method ensures that higher performing annotators contribute more to the final landmark position which is in contrast to conventional methods that assume all annotators are equally accurate in completing the set task. In addition to describing the novel method, a set of preliminary experimental results is also provided, comparing the performance of the precision method to that of the global mean.

Automatic Nasal PAP Mask Sizing with a Deep Unet

We present the use of a deep Unet convolutional neural network as an automated way of sizing nasal Positive Airway Pressure (PAP) masks using facial images of patients. Using a VGG16 backbone the network was trained with the MUCT dataset and a significant amount of data augmentation. The trained model was then applied to a small custom dataset of PAP and non-PAP patients to predict the nose widths and corresponding PAP mask sizes of each subject. The Unet model produced a mask sizing accuracy of 63.73% (116/183) and a within one size accuracy of 88.5% (162/183).

The Effect of Landmark Variability on Automated PAP Mask Sizing

We present a study of the effect of variability in manual facial landmarking annotation on image based positive airway pressure mask sizing. Employing 12 expert annotators to identify specific facial landmarks on two images, ANOVA analysis was used to investigate landmark variability attributed to the annotator, types of landmarks and images. We found that while there was agreement between annotators, variance in individual landmark positions resulted in an error of up to 13.59% in measurements defining mask size. A number of different options are presented for reducing errors from manual facial landmarking.

Automatic PAP Mask Sizing with an Error Correcting Autoencoder

We present the use of an error correcting autoencoder stage to a convolutional neural network model as a means of improving image based automatic Positive Airway Pressure (PAP) mask sizing accuracy. A single convolutional layer neural network was pre-trained using MUCT dataset and transfer learning was applied to mitigate against the relatively small custom dataset. The base model was then augmented with an additional error correcting autoencoder and trained against the custom dataset. The presented model increased PAP sizing accuracy against the baseline by 15.3% while reducing overfitting.