Deep Learning based detection of hypoxic-ischemic encephalopathy
In cooperation with a few friends from different departments at the University Hospital Charité in Berlin, I started in 2022 to work on a software focused on the development of an advanced deep-learning model for the detection of hypoxic-ischemic encephalopathy (HIE) after cardiac arrest on CT scans. We aim to revolutionize the diagnosis and characterization of HIE through the application of artificial intelligence.
In our first study, we conducted a comprehensive retrospective analysis of head CT scans obtained from patients who experienced cardiac arrest. The scans were meticulously classified into two categories: severe HIE and no signs of HIE. To train and validate our deep learning models, we employed a division of the images into a training set and a test set. Multiple deep learning models, based on Densely Connected Convolutional Networks, were evaluated using different image input formats, including both 2D and 3D representations.
The results of our research were highly promising, with all optimized networks exhibiting the ability to detect signs of HIE. Notably, our models based on 2D image stacks demonstrated particularly encouraging performance metrics. We also incorporated visual explainability data using Gradient-weighted Class Activation Mapping to enhance transparency and provide insights into the decision-making process of our AI model.
This proof-of-concept deep learning model represents a significant advancement in the field of HIE detection. By accurately identifying signs of HIE on CT images, our research has the potential to significantly impact clinical practice. The exceptional performance of our 2D image stack models suggests exciting prospects for the implementation of this technology in routine clinical settings, empowering healthcare professionals to enhance their characterization of imaging data and make more informed predictions regarding patient outcomes.
As we continue to validate our findings and refine our deep learning model, we foresee a future where CT-based HIE detection becomes an integral part of clinical routine. Our research has the potential to change the diagnostic landscape, providing healthcare professionals with advanced tools to improve patient care and prognosis.
I successfully finished this exciting research project with the publication and defense of my doctoral thesis in 2025.
Whole-Body Hyperthermia
TBD
EWOT
TBD