Deep Health Unit
The Deep Health Unit (DHU) is the research group oriented to the development of Artificial Intelligence (AI) pmodels and algorithms for decision support in the clinical field.
Bayesian probabilistic models, deep learning algorithms and ensemble algorithms are designed and developed for analysing multimodal data (structured tabular data, images, unstructured text, time series), even in large quantities, according to big data analytics approaches.
Areas of Interest
- Deep neural networks for:
- automatic extraction of features from multimodal imaging (AI-based biomarkers)
- development of decision support models in customised medicine (diagnostic and prognostic risk models)
- automatic segmentation of multimodal radiological images (MRI, CT, pet) for rapid and accurate extraction of quantitative clinical features
- Structuring of free texts from medical records in Italian, using Natural Language Processing (NLP) approaches
- Probabilistic programming for the implementation of Bayesian models
- Ensemble Algorithms
- Causal inference from observational data
The collaboration between DHU and the Epidemiology of the Institute of Clinical Physiology of the CNR led to the establishment of the DataLearnLab (D2L) laboratory.
- Ripoli A, Sozio E, Sbrana F et al. Personalized machine learning approach to predict candidemia in medical wards. Infection 2020; 48: 749-759.
- Salvatori M, Martini N, Ripoli A et al. A generalised deep learning approach for the quantification of the hepatic fat by non-contrast CT imaging. ECR 2020. doi: 10.26044/ecr2020/C-04146.
- Della Latta D, Martini N, Aztori S. A deep learning approach to estimate three-dimensional FGT maps from multimodal MRI images. ECR 2020; doi: 10.26044/ecr2020/C-11630.
- Martini N, Fabbri C, Ripoli A et al. Estimation of cardiac aging by CT imaging: towards the definition of new cardiovascular biomarkers using a deep learning approach. ECR 2020. doi: 10.26044/ecr2020/C-11638.
- Margaryan R, Della Latta D, Bianchi G et al. Intercostal space prediction using deep learning in fully endoscopic mitral valve surgery. ISMICS Annual Scientific Meeting 2019 (Subramanian Innovation Award).
- Martini N, Della Latta D, Positano V et al. T2* mapping by segmental approach using deep learning. SCMR 22nd Annual scientific Session. Seattle, 2019.
- Martini N, Vatti A, Ripoli A et al. Robust reconstruction of cardiac T1 maps using recurrent neural networks. International Conference on Medical Imaging with Deep Learning. London, 2019. https://arxiv.org/abs/1907.12454.
- Della Latta D, Santing G, Martini N et al. Dual output V-Net CNN: a virtual iodinated contrast mediainjection in chest CT toward a new cardiac risk assessment. RSNA 2019; AI022-EB-MOB.
- Della Latta D, Santini G, Valvano G. Contrast-free estimation of cardiac volumes from CT scans using deep learning. ECR 2018; doi: 10.1594/ecr2018/C-1413.
- Valvano G, Santini G, Martini N et al. Convolutional neural networks for the segmentation of microcalcification in mammography imaging. J Healthcare Enigneering 2018; doi: 10.1155/2019/9360941.
- Della Latta D, Santini G, Martini N et al. Deep learning for discovery of latent information in contrast-free cardiac CT images. RSNA 2018.
- Ripoli A, Della Latta D, Martini N. Radiophenomics: a machine learning approach to radiological exploration of atherosclerosis. RSNA 2017.
- Ripoli A, Rainaldi G, Rizzo M, Mercatanti A, Pitto L. The fuzzy logic of microRNA regulation: a key to control cell complexity. Curr Genomics 2010; 11:350-353.
- Meloni A, Ripoli A, Positano V, Landini L. Mutual information preconditioning improves structure learning of Bayesian networks from medical databases. IEEE Trans Inf Technol Biomed 2009; 13: 984-989.
Gilead Digital Health Programme 2018: “Development of artificial intelligence to support clinical decision making in cases of candidaemia or suspected invasive candidiasis.”