Predicting the severity of obstructive sleep apnea based on paranasal sinus computed tomography scan using multimodal deep learning models

Obstructive Sleep Apnea (OSA) is a common sleep disorder that can lead to serious health problems if left untreated. Paranasal sinus computed tomography (CT) scans have shown promise in predicting OSA severity, but manual interpretation is time-consuming and error-prone.

This project implements multimodal deep learning models to predict the severity of OSA based on paranasal sinus CT scans. The models take both structured and unstructured data as inputs, and are trained and tested using the provided dataset.

Requirements

Package	Version
Python	>= 3.6 (3.10 recommended)
TensorFlow	>= 2.8.0 (2.9.0 recommended)
OpenCV	Lastest
tqdm	Lastest
Pandas	Lastest
PyDICOM	Lastest

To install the required packages, run the following command:

pip install -r requirements.txt

Usage

Training

To train the models, run the following command:

python train.py --input_shape_structured 3 --input_shape_unstructured None,512,512,1 --batch_size 32 --epochs 10 --validation_split 0.2 --save_weights --save_weights_path ./model_weights.h5 --load_weights --load_weights_path ./model_weights.h5 --dataset_path ./dataset.npz

The arguments are:

Argument	Description
input_shape_structured	Input shape for structured data (comma-separated).
input_shape_unstructured	Input shape for unstructured data (comma-separated; D, H, W, C).
batch_size	Batch size for training.
epochs	Number of epochs for training.
validation_split	Fraction of the training data to be used as validation data.
save_weights	Whether to save the model weights after training.
save_weights_path	Path for saving model weights.
load_weights	Whether to load model weights before training.
load_weights_path	Path for loading model weights.
dataset_path	Path for the dataset.

Testing

To test the models, run the following command:

python test.py --input_shape_structured 3 --input_shape_unstructured None,512,512,1 --weights_path ./model_weights.h5 --test_dataset_path ./test_dataset.npz

The arguments are:

Argument	Description
input_shape_structured	Input shape for structured data (comma-separated).
input_shape_unstructured	Input shape for unstructured data (comma-separated; D, H, W, C).
weights_path	Path for loading the trained weights.
test_dataset_path	Path for loading the test dataset.

Data Availability

The data that support the findings of this study are available from the corresponding author, Jin Youp Kim (kjyoup0622@gmail.com), upon reasonable request.