AI-driven data analysis with proprietary machine-learning methodology
Most analytical pipelines give an answer without saying how sure they are. We analyse clinical, imaging, omics and registry data with proprietary methodology built on Professor Arkadiusz Sitek's work on Bayesian computing and machine learning: origin-ensemble and Bayesian estimation for noisy measurements, factor analysis of dynamic signals, and transformer-based models of health trajectories. Every result comes with its uncertainty, its validation plan and a plain statement of what the data cannot support.
Typical questions
- Can an existing dataset answer the question we are about to run a new study for?
- How reliable is a prediction model, in our population, before we act on it?
- Which signals in imaging or time-series data carry information, and which are noise?
- Can a foundation model of health records be used responsibly for risk estimation in our setting?
What you receive
- 1An analysis plan: the question, a data audit, pre-specified models, the validation strategy and what will count as a negative result
- 2The analysis itself: Bayesian or machine-learning models fitted to your data, with calibration, uncertainty and subgroup performance reported, and code your team can run
- 3A written report with figures, limitations and recommendations, and where wanted a methods description suitable for a scientific report
Well suited to
Biotech and pharmaceutical teams with trial or real-world data, imaging and diagnostics companies, academic groups with biobank or registry access, and investors who need an independent reading of a model's performance claims.
Analyses run on data you are entitled to share, under a written agreement, within the EEA. We report what the data show, including null results, and do not provide medical, regulatory or legal advice.
Selected publications behind the method
The algorithms, in the primary literature.
The methodology draws on Professor Sitek's published work in Bayesian computing, statistical image reconstruction and machine learning for clinical data. Each entry links to the original publication.
- 1Reconstruction of emission tomography data using origin ensembles (opens in a new tab)
Sitek A. IEEE Trans Med Imaging. 2011;30(5):946-956.
- 2Data analysis in emission tomography using emission-count posteriors (opens in a new tab)
Sitek A. Phys Med Biol. 2012;57(21):6779-6795.
- 3Validation of Bayesian analysis of compartmental kinetic models in medical imaging (opens in a new tab)
Sitek A, Li Q, El Fakhri G, Alpert NM. Phys Med. 2016;32(10):1252-1258.
- 4Statistical Computing in Nuclear Imaging (opens in a new tab)
Sitek A. CRC Press, Series in Medical Physics and Biomedical Engineering; 2014.
- 5Zero shot health trajectory prediction using transformer (opens in a new tab)
Renc P, Jia Y, Samir AE, et al., Sitek A. npj Digit Med. 2024;7:256.
- 6Foundation model of electronic medical records for adaptive risk estimation (opens in a new tab)
Renc P, et al., Sitek A. GigaScience. 2025;14:giaf107.
- 7Direct estimation of fetal biometry measurements from ultrasound video scans through deep learning (opens in a new tab)
Płotka S, et al., Sitek A. Am J Obstet Gynecol MFM. 2025;7:101623.
- 8Vascular phenotypes in early hypertension (opens in a new tab)
Murray EC, Delles C, Orzechowski P, Renc P, Sitek A, Wagenaar J, Guzik TJ. J Hum Hypertens. 2023;37:898-906.
- 9Beyond language: generative artificial intelligence as a general computing model for medicine (opens in a new tab)
Sitek A, Bates DW. Lancet Digit Health. 2026;8:101011.