Research
Renal outcomes, multimodal learning, clinical NLP, robotics vision and registry-scale research.
On this page
Journal articles
4
Full journal articles, counted separately from conference abstracts and preprints.
Public preprints
4
Public manuscript versions, including earlier versions of published journal work.
Conference research records
13
Published abstracts and supervised research records, with roles and listing status documented.
Talk and poster records
17
Personal talks, collaborator/student presentations and explicitly labelled poster listings.
Themes
Research themes
Theme 01
Clinical AI systems with deployment discipline
Building models that remain useful after contact with workflow, missingness, calibration drift, and resource constraints.
- Prospective framing over leaderboard thinking
- Transparent evaluation and clinically faithful labels
- Models shaped by bedside usefulness, not only benchmark lift
Theme 02
Renal outcomes and nephrology decision support
Current doctoral work focuses on fair, interpretable prediction of renal outcomes with attention to equity, lead time, and actionability.
- Acute kidney injury forecasting
- Risk stratification for heart failure and renal trajectories
- Interpretability aligned with clinician reasoning
Theme 03
Multimodal modeling and MMLM post-training
Post-training multimodal models with attention to grounded reasoning, evaluation quality, and robustness before deployment.
- Post-training and evaluation of multimodal systems
- Fusion across signals, text, and imaging
- Reasoning quality measured before deployment
Theme 04
Clinical NLP, agentic systems, and robotics perception
Using LLM-assisted extraction, tool-using agents, and vision systems without losing auditability, safety, or translational usefulness.
- Clinical note extraction with verification loops
- Agent patterns tied to auditability and validation
- Computer vision within robotics and embodied workflows
Programs
Current programs
Doctoral focus
Renal outcome prediction with clinically usable lead time
Doctoral work on prediction of renal outcomes with an emphasis on fairness, interpretability, and decisions that can be acted on in practice.
Methods
temporal modeling / fairness evaluation / interpretability analysis
Metrics
AUROC / calibration / lead-time utility
Datasets
nephrology cohorts / longitudinal clinical records
Post-training
MMLM post-training for clinical reasoning
Post-training multimodal models with attention to grounded reasoning, evaluation quality, and deployment-minded reliability.
Methods
post-training / evaluation design / multimodal alignment
Metrics
task accuracy / groundedness / failure analysis
Datasets
multimodal medical corpora / instruction datasets
Registry-scale work
Liver transplant registry modeling at EU scale
Research infrastructure and modeling on the largest single-institute liver transplant registry in Europe, built for rigorous translational analysis.
Methods
registry curation / outcome modeling / cohort design
Metrics
cohort completeness / discrimination / clinical interpretability
Datasets
liver transplant registry / linked clinical records
Papers, preprints and conference research
Peer-reviewed journal articles and accepted manuscripts.
4 entries visible in publications.
Forecasting ICU Acute Kidney Injury with Actionable Lead Time Using Interpretable Machine Learning: Development and Multi-Center Validation
Extends kidney injury prediction toward real lead-time utility instead of retrospective signal mining alone.
Summary
Interpretable machine-learning models are developed to forecast ICU acute kidney injury with actionable lead time, including multi-center validation to test generalizability.
Abdulla Zahi Hourani; Zuzanna Jakubowska; Jolanta Malyzsko
Prognostic Value of Different Iron Status Definitions in Congestive Heart Failure: A Retrospective MIMIC-IV Analysis of Risk Stratification and Mortality
Shows how more precise phenotyping can materially change downstream prognostic separation.
Summary
Retrospective MIMIC-IV analysis compares multiple iron-status definitions for one-year mortality risk stratification in congestive heart failure.
Hourani A; Surmeli A; Devarapalli S
Clinical Picture and Outcomes in Patients Diagnosed with Brain Abscess
Pairs clinically grounded infection research with a stronger systems view of hospitalization trajectories.
Summary
Retrospective cohort study characterizes the presentation and outcomes of patients with brain abscess, highlighting factors linked to hospital course and prognosis.
Furman-Dlubala A; Bednarska A; Radkowski M; Paciorek M; Kolodziejska J; Laskus T; Bursa D; Porowski D; Makowiecki M; Hourani A et al.
The Impact of COVID-19 Non-Pharmaceutical Interventions on Notifiable Infectious Diseases in Poland: A Comprehensive Analysis from 2014-2022
Connects model thinking to surveillance, policy shifts, and long-horizon epidemiologic interpretation.
Summary
A population-level analysis of notifiable infectious disease trends in Poland from 2014 to 2022, covering the period shaped by COVID-19 interventions.
Abdulla Zahi Hourani; Abdelrahman Abdelsalam; Arman David Surmeli
Impact
Outputs, recognition, and infrastructure.
Awards
Repeated placements across clinical and AI conference sessions
Conference work has produced first-, second-, and third-place results, plus nephrology recognition, suggesting strong translational framing as well as technical depth.
Doctoral focus
Fair and transparent renal outcome prediction
Current PhD work centers on interpretable risk models for nephrology with attention to equity-aware evaluation and clinically faithful stratification.
Infrastructure
Project Quintessence and research infrastructure
Current work extends beyond papers into research infrastructure, registry-scale collaboration, multimodal systems, robotics, and open technical builds.
Methods stack
- Interpretable machine learning
- Temporal forecasting and lead-time evaluation
- MMLM post-training and multimodal evaluation
- Clinical NLP and LLM-assisted extraction
- Registry-scale cohort design and data curation
- Calibration, external validation, and failure analysis
- Computer vision and agentic workflow design
- Human-in-the-loop review for safety-sensitive outputs
Institutions and partners
- Medical University of Warsaw
- European Renal Association
- Khalifa University
- Project Quintessence collaborators
- Clinical and industry research partners
Contact
Research collaborations work best when the clinical question and the technical execution are both taken seriously.
If the overlap is prediction, post-training, registry science, clinical NLP, or translational evaluation, that is likely a good fit.
On this page