1. A Guide to 'Registry Science' for Data Leaders

Understanding how different populations respond to diseases, treatments, and therapies is one of medicine’s biggest challenges. Tracking patient groups over time to study disease progression is especially tough — particularly for rare diseases. 

One approach is to use data generated during routine healthcare, like medical records and pharmacy claims, often called “real-world evidence” (RWE). However, this data comes with challenges, including inconsistencies, missing information, and system incompatibilities, requiring thorough management for research.

_{Image Source: https://academic.oup.com/jamia/article/31/4/1001/7613411}

With over 40 years of experience as informaticians, the Quantori authors have worked on countless registry projects and data.  

In this article, they share insights to help clarify concepts that aren’t always well-understood in the industry.

Quantori author: Steven Labkoff, MD, former Quantori Global Head of Registry Science.

2. Explainable AI for Spotting Pulmonary Edema in X-rays

Pulmonary edema is a serious, potentially life-threatening condition and a leading cause of hospitalization for patients with congestive heart failure. When patients are admitted to the hospital, assessing the severity of pulmonary edema is key to choosing the right treatment. Radiographic imaging is essential for this evaluation and is often used to track how the condition progresses.

_{Image Source: https://academic.oup.com/radadv/article/1/1/umae003/7630768}

The article introduced a novel deep-learning methodology that assists in diagnosing and classifying the severity of pulmonary edema from chest X-rays. This study is the initial step toward creating a fully automated tool to help clinicians diagnose and assess the severity of pulmonary edema.

Quantori authors: Viacheslav Danilov, PhD, Anton Makoveev, PhD, Alex Proutski, PhD, Irina Ryndova, Yuriy Gankin, PhD.

3. Improving Machine Learning in Microbiomics 

Machine learning (ML) has the potential to transform clinical microbiomics, especially in disease diagnosis and prediction. However, for ML to succeed in these areas, models must be reproducible, interpretable, and meet strict regulatory standards. This study highlights key improvements needed to align current ML practices with clinical requirements. To do so, scientists analyzed 100 peer-reviewed articles from 2021 to 2022. The study shares tips to avoid common ML pitfalls in microbiomics and includes an interactive tutorial to support best practices.

_{Image Source: https://www.sciencedirect.com/science/article/pii/S2666827024000835}

Quantori authors: Natasha Dudek, PhD, Mariami Chakhvadze, Saba Kobakhidze, Omar Kantidze, PhD, Yuriy Gankin, PhD.

4. A Machine Learning Model for Understanding Chemical Structures

Identifying a compound's chemical structure is a critical task in chemistry, and Nuclear magnetic resonance spectroscopy (NMR) remains one of the most powerful techniques for this. This article introduces the Structure Seer model, a machine-learning approach that predicts atom connectivity in molecules based on a molecule's elemental composition from NMR spectra.

_{Image Source: https://www.sciencedirect.com/org/science/article/pii/S2635098X24000032}

This approach has great potential for scalability, as it can leverage large amounts of data on known chemical structures for the model’s learning.

Quantori author: Denis Sapegin, PhD.

5. Responsible AI in Clinical Decision Support

Integrating AI in healthcare can improve clinical decision-making, but challenges like ensuring trust, reducing bias, and ensuring safety must be addressed. The lack of clear methods for evaluating AI tools before and after deployment, especially for transparency, performance, and adverse event reporting, makes this situation challenging.

_{Image Source: https://academic.oup.com/jamia/article/31/11/2730/7776823}

This paper aims to provide practical suggestions for creating methods, rules, and guidelines to ensure that the development, testing, supervision, and use of AI in clinical decision support systems are done well and safely for patients.

Quantori author: Steven Labkoff, MD, former Quantori Global Head of Registry Science.

Explore more of Quantori's academic work and innovations here: Publications & White Papers | Quantori.