Giorgio Ricciardiello
Building ML systems that turn physiological signals into clinical insight
Building ML systems that turn physiological signals into clinical insight
I'm a Machine Learning Scientist and Biomedical Researcher with 5+ years building multimodal AI and signal-processing systems for neural, wearable, and clinical data. My pipelines span cohorts exceeding 2 million patients — from actigraphy-derived sleep biomarkers that predict Parkinson's disease a decade before diagnosis, to population-scale causal inference for stroke and dementia prevention.
Currently a Research Scientist at Mount Sinai. Previously a Research Assistant in the Mignot Laboratory at Stanford. I hold an MS in Epidemiology & Clinical Research from Stanford, an MS in Biomedical Engineering from the Technical University of Denmark, and a BS in Biomedical Engineering from the University of Genoa.
Multimodal ML pipelines integrating wearable sensor data and EHR outcomes. Causal inference platform analyzing stroke and dementia prevention strategies at population scale using AIPW, TMLE, and marginal structural models.
ML diagnostic classifier for Narcolepsy Type 1. Large-scale benchmarking across 69,000 polysomnography records. Conv-LSTM and CNN models for automated REM Sleep Behavior Disorder detection with cross-cohort generalization.
Real-time PPG signal-processing algorithms for commercial wearable devices. Validation protocols supporting ISO 13485 and EU MDR regulatory submissions.
Contributed to a multinational consortium on digital biomarkers and ML approaches for sleep disorders.
Three proof points of solving difficult technical problems at scale—published, externally validated, and reproducible.
End-to-end machine learning on 87,510 UK Biobank participants
Built an end-to-end machine learning and survival modeling pipeline to identify early markers of Parkinson's disease using 7-night wrist accelerometry from the UK Biobank. I engineered the full Python framework—including feature engineering, Cox survival analysis, inverse probability weighting, SHAP explainability, and reproducible pipelines—to detect disease years before clinical diagnosis. The model identified individuals with a 4.69× higher risk of Parkinson's disease over ~10 years, establishing scalable wearable biomarkers for neurodegenerative disease.
$250,000 ResMed grant for objective Narcolepsy screening
Secured a $250,000 competitive research grant from ResMed to develop machine learning methods for objective Narcolepsy type 1 screening. I led the machine learning methodology, implementing nested cross-validation and Bayesian hyperparameter optimization to build models achieving an AUC of 0.996 and 99.2% specificity. The work was published in the peer-reviewed journal Sleep, demonstrating successful translation of AI research into clinically applicable diagnostics.
Automated wound-healing quantification without annotated data
Developed a novel two-stage computer vision pipeline for automated wound-healing quantification that combined variance-based segmentation with temporal continuity modeling, eliminating the need for annotated training data. I engineered the complete Python framework with quality control, parallel processing, and extraction of biologically meaningful metrics across 134 microscopy experiments. The pipeline substantially reduced manual analysis while enabling reproducible, high-throughput phenotypic screening for biomedical research.
Open to research collaborations, clinical AI consulting, and data science roles at the intersection of medicine and machine learning. Reach out on LinkedIn.