AI and quantum computing
Quantitative radiomic feature extraction from medical images
Radiomics refers to the process of extracting numerical features from images. Radiomics combined with PET and hybrid imaging is a particularly important approach in the field of in vivo tumor characterization. Radiomics has challenges given its sensitive nature to patient preparation, imaging and reconstruction and feature extraction parameter variations. Here, we focus on establishing radiomic feature extraction approaches that minimize such variations in radiomic models across different imaging centers.
Data preprocessing optimization
Most medical datasets are inherently imbalanced due to the fact that the occurrence of disease subgroups – especially in case of rare diseases – is imbalanced. Furthermore, data may contain noisy or redundant features as well as outliers. Besides the above issues, high-dimensionality of data further challenge the process of machine learning (ML). Current approaches to purify and prepare data for ML are purely mathematical approaches, not considering human expert (physician, physicist) knowledge. Here, we focus on the establishment of a formal language to encode human knowledge in order to guide an AI-driven automated data preprocessing pipeline establishment process to maximize ML performance.
Holomics and ensemble learning
Holomics refers to holistic multi-omics and it is a process which combines imaging and non-imaging (e.g. clinical and patient demographics) data. Holomics challenges ML processes, as their complexity may magnify ML training issues such as bias-variance tradeoff, curse of dimensionality as well as method bias. Ensemble learning establishes various, heterogeneous machine learning and deep learning predictive models and combines them to a mixed model to minimize the effect of the above issues. Here, we focus on the automated establishment of mixed-stacked ensemble learners (a.k.a. super learners) in combination with AI-driven data preprocessing pipelines to support ML predictive learning processes.
Quantum computing
Quantum computing is a relatively new field of computational science which aims to revolutionize how we process data and solve problems that are NP-complex. Here, we focus on the establishment of high-level quantum AI algorithms, quantum ensemble learning, and on a broader scale, quantum optimization to support the establishment of future modeling, simulation, reconstruction and AI approaches conducted at our center.
Research Groups
Highlights
“Images are data, and data is knowledge.” This comprehensive introduction of the concept of holomics as the next frontier in clinically relevant data sciences was published by Laszlo Papp and colleagues in open access.