CCTA scenario

Background: Addressing Coronary Artery Disease (CAD) early is crucial for reducing heart disease’s global impact. The demand for non-invasive, cost-effective diagnostic tools is rising, driven by the need to minimize patient risk and healthcare costs.

Problems: Traditional CAD diagnostics, often invasive and reliant on contrast agents, present safety and cost concerns. There’s a critical need for innovative methods that leverage patient data without these drawbacks.

Objectives: Develop an AI-powered, non-invasive diagnostic application that:

  • Analyzes lab reports, calcium scoring, and epicardial fat volume to predict obstructive CAD risk.
  • Offers a safer, cost-effective alternative to contrast-enhanced CCTA.
  • Supports early intervention with accurate, timely risk assessment.

 

The Solution: An AI application utilizing machine learning to predict CAD from non-contrast data, including laboratory findings and imaging results. It aims for high accuracy without immediate CCTA use, prioritizing patient safety and reducing costs.

SHapley Additive exPlanations (SHAP) enhance model interpretability, giving clinicians insight into predictions.

 

 

Value Proposition:

  • Reduces healthcare costs and improves access by eliminating the need for contrast agents and invasive tests.
  • Enhances patient safety, particularly for those at risk from contrast media.
  • Optimizes healthcare resource use, reserving more invasive tests for high-risk patients.
  • Seamlessly integrates into clinical workflows, providing evidence-based insights for decision-making.
  • Improves patient outcomes with early risk detection and targeted interventions.

This AI tool transforms CAD risk assessment, offering a non-invasive, efficient pathway to early detection and management, aligning with modern healthcare objectives to improve care quality and reduce unnecessary procedures.

 

Obstetrics scenario

Background: In obstetric care, accurately predicting preterm birth (PTB) and adverse neonatal outcomes resulting from Fetal Growth Restriction (FGR) is crucial for improving neonatal and maternal health outcomes. Leveraging Artificial Intelligence (AI) and Machine Learning (ML), new predictive tools have been developed to forecast these risks with unparalleled accuracy, utilizing complex datasets.

Problems: Traditional PTB prediction methods have been constrained by a reliance on a limited set of risk factors, failing to fully account for the complex causes of preterm delivery. Similarly, predicting specific adverse outcomes from FGR has been challenging due to the reliance on indirect fetal well-being assessments and maternal risk factors, often missing subtle risk indicators.

Objectives: The overarching goal of these AI-based applications is to:

1) Analyze extensive datasets to forecast risks with high precision. 2) Employ sophisticated ML algorithms, including an Optimized Voting ensemble model for PTB and Random Forest for FGR, enhancing accuracy and reliability. 3) Prioritize high recall for PTB predictions to minimize the risk of overlooking at-risk pregnancies. 4)Provide interpretable predictions through Shapley Additive Explanations (SHAP), building clinician trust.

The Solution: Two advanced AI-based applications have been developed:

  • PTB Application: Utilizes an extensive dataset to predict PTB with an Optimized Voting model, achieving a recall value of 0.97, ensuring nearly all true PTB cases are accurately identified.
  • FGR Application: Targets predicting specific adverse neonatal outcomes from FGR using a sophisticated dataset, employing the Random Forest model for high-risk identification.

Both applications combine accuracy with interpretability, offering clinicians transparent insights into the prediction process.

Value Proposition: These AI-based tools offer significant advantages, including:

  • For PTB: Dramatically improving the identification of high-risk pregnancies, enabling early interventions.
  • For FGR: Significantly improving the prediction of adverse neonatal outcomes, allowing for preemptive management strategies.
  • Both applications reduce dependence on subjective assessments and enhance prenatal care efficiency, potentially lowering healthcare costs.
  • By providing comprehensive, data-driven analyses and clear, interpretable predictions, these tools foster clinician confidence in AI technologies, marking a significant leap forward in obstetric care.

These AI-driven applications represent a novel, evidence-based approach to enhancing outcomes in obstetric care, targeting both PTB and adverse outcomes from FGR with high precision and reliability.

 

 

VCE scenario

Background: Misdiagnosis or delayed diagnosis are two of the most common types of medical malpractice in gastroenterology. They often result in patients not receiving proper and timely care, potentially followed by a serious deterioration of their health or even death.

Problems: Capsule endoscopy is the only and very expensive technique to identify and examine the small bowel. In clinical practice, the use of video capsule endoscopy is a very time-consuming and tedious task. The user, typically a gastroenterologist, first needs to watch a very long video (duration > 1 hour) and then to analyse the findings.

Objectives: Develop an AI-powered web platform that:

  • Automatic detects of potential abnormalities.
  • Allows capsule endoscopy video reading similar to current practice.
  • Improves diagnostic performance.
  • Reduces the duration of the examination.

The Solution: A web-based solution powered with AI that accelerates the capsule endoscopy reading process. The main functionalities include

  • Uploading of capsule endoscopy videos to the web application
  • Applying an AI pipeline to extract frames with suspicious findings and presenting the outcomes in a user-friendly manner,
  • Allowing the doctor to interact with the AI findings, annotate new findings, insert comments and make the final diagnosis.
  • All information created by the reviewer, i.e., comments, annotations, and the final diagnosis, are saved in the application for future reference.

Value Proposition: The whole process of reviewing a capsule endoscopy video has been enhanced and optimised via allowing the collaboration of the doctor with the AI.

 

ECHO scenario

Background: The assessment of left ventricular (LV) function is important for diagnosis, management, follow-up, and prognostic evaluation of patients with heart problems. One critical task in analysing LV function is to accurately quantify ejection fraction (EF) and global longitudinal strain (GLS) with echocardiography (ECHO).

Problems: An important procedural problem that appears in quantification of LV-EF and LV-GLS is the inconclusive decision due to the subjectivity of clinician’s interpretation of echocardiogram clip that could impact clinical care. Moreover, the experience of the clinicians plays a significant role in the interpretation of measurement. All in all, interobserver variability can cause interpretation conflicts, delays, and unnecessary follow-up meetings which burden the clinical routine of an ECHO laboratory.

Objectives: Develop a computer aided diagnosis web application that:

  • Automatically processes ECHO clips to calculate LV-EF and LV-GLS.
  • Offers visual explanations regarding the areas within the ECHO frames that are involved in the calculations.
  • Supports accurate, timely and observer-independent LV assessment.

The Solution: The echocardiography tool (ECHO tool) is a computer-aided diagnosis software application that assists cardiologists to assess LV function. The ECHO tool is a web application via which the end-user, i.e., a cardiologist, can 1) import echocardiographic data, i.e., multi-frame DICOM files, 2) navigate and study the imported data, 3) apply AI algorithms to automatically estimate LV function related parameters, i.e., LV-EF and LV-GLS, and 4) save the AI outcomes along with relevant metadata.

Value Proposition: Improving the echocardiography laboratory services via reducing the interobserver variability in assessing cardiac function.

Main solution providers:

Aristotle University of Thessaloniki (AUTH)

Prof. Panagiotis Bamidis (bamidis@med.auth.gr), Prof. Leontios Hadjileontiadis (leontios@auth.gr)

Hosting Facilities:

AHEPA University General Hospital

Assoc. Prof. George Giannakoulas (ggiannakoulas@auth.gr) (CCTA)

Assoc. Georgios Germanidis (geogerm@auth.gr) (VCE)

Ippokrateio General Hospital of Thessaloniki

Prof. Vasileios Vassilikos (vvassil@auth.gr) (ECHO)

Assist. Prof. Dr. Themistoklis Dagklis themistoklis.dagklis@gmail.com  (Obstetrics)

Tags: