arXiv subjects

Robin Andlauer

Publications and source records attributed to Robin Andlauer.

2 recordsLinked to original sources

arXiv2021

Prediction of mandibular ORN incidence from 3D radiation dose distribution maps using deep learning

Background. Absorbed radiation dose to the mandible is an important risk factor in the development of mandibular osteoradionecrosis (ORN) in head and neck cancer (HNC) patients treated with radiotherapy (RT). The prediction of mandibular ORN may not only guide the RT treatment planning optimisation process but also identify which patients would benefit from a closer follow-up post-RT for an early diagnosis and intervention of ORN. Existing mandibular ORN prediction models are based on dose-volume histogram (DVH) metrics that omit the spatial localisation and dose gradient and direction information provided by the clinical mandible radiation dose distribution maps. Methods. We propose the use of a binary classification 3D DenseNet121 to extract the relevant dosimetric information directly from the 3D mandible radiation dose distribution maps and predict the incidence of ORN. We compare the results to a Random Forest ensemble with DVH-based parameters. Results. The 3D DenseNet121 model was able to discriminate ORN vs. non-ORN cases with an average AUC of 0.71 (0.64-0.79), compared to 0.65 (0.57-0.73) for the RF model. Conclusion. Obtaining the dosimetric information directly from the clinical radiation dose distribution maps may enhance the performance and functionality of ORN normal tissue complication probability (NTCP) models.

physics.med-ph↗

arXiv2021

Uncertainty-Aware Training for Cardiac Resynchronisation Therapy Response Prediction

Evaluation of predictive deep learning (DL) models beyond conventional performance metrics has become increasingly important for applications in sensitive environments like healthcare. Such models might have the capability to encode and analyse large sets of data but they often lack comprehensive interpretability methods, preventing clinical trust in predictive outcomes. Quantifying uncertainty of a prediction is one way to provide such interpretability and promote trust. However, relatively little attention has been paid to how to include such requirements into the training of the model. In this paper we: (i) quantify the data (aleatoric) and model (epistemic) uncertainty of a DL model for Cardiac Resynchronisation Therapy response prediction from cardiac magnetic resonance images, and (ii) propose and perform a preliminary investigation of an uncertainty-aware loss function that can be used to retrain an existing DL image-based classification model to encourage confidence in correct predictions and reduce confidence in incorrect predictions. Our initial results are promising, showing a significant increase in the (epistemic) confidence of true positive predictions, with some evidence of a reduction in false negative confidence.

eess.IV↗