‍

Why did we choose CVD?

‍

Challenge

Despite technological advancements, especially in India current predictive models for CVDs often lack transparency, making it difficult for medical practitioners to trust and understand their predictions.

Need

Innovative solutions that provide high accuracy and clear explanations for decision-making in CVD prediction.

‍

Research Focus

Objective: Develop an accurate and interpretable predictive model for CVDs using machine learning and XAI techniques.

‍

Methodology

‍

Attributes: 11 demographic, clinical, and diagnostic parameters.

‍‍

Finding the best model and dataset combination

‍

ExplainableAI (XAI) Techniques

‍

Local Interpretable Model-agnostic Explanations (LIME)

‍

SHapley Additive exPlanations (SHAP)

‍

‍

‍

‍

What does our UI display? How would it help medical practitioners?

Our implementation offers a user-friendly interface designed to assist medical practitioners in heart disease prediction. By inputting 11 key parameters such as age, sex, and cholesterol levels, the model processes the data to predict the likelihood of heart disease, providing valuable insights to support clinical decision-making. The tool enhances transparency through explainability features, including LIME explanations and SHAP plots, which clarify individual predictions and overall model behavior. This empowers doctors with a deeper understanding of the factors influencing predictions, improving the accuracy and confidence of heart disease assessments.

For fields like Chest pain, or other categorical input value parameters, we have made a drop-down menu to not avoid complications with such parameters. Final call is of course taken by the medical practitioner to validate the output.

Future scope

‍

Takeaways

Refer to our research paper published in the 3rd Annual IEEE Conference here.

‍

‍