In this study we develop an artificial intelligence (AI) framework to deliver a 3D anatomical lung model with the patient's disease profile and the uncertainty estimation of the diagnosis. We study a multi-classification problem of pulmonary hypertension diseases using different deep learning models. We estimate the epistemic and aleatoric uncertainty of the framework and datasets, and we used generalised explainable and feature engineering techniques (PCA, GradCam) to study and validate the classification predictions of the deep learning network. The AI framework was evaluated in an unbiased validation scheme (internal / unseen datasets) delivered accurate, robust and generalised results.
Please if you uce the code cite bellow studies:
cd ./PH_3Dpatches-master
pip install .
cd ./PH_3Dpatches-master/script/
python3 train_patch.py train_patch.config