# Multi-phase deep learning model for automated disease classification from cardiac cine MRI

## TL;DR

This paper presents a deep learning model that automatically classifies cardiovascular diseases from heart MRI scans and explains its decisions.

## Contribution

A novel multi-phase deep learning model that improves classification accuracy and provides model explainability for cardiac MRI-based disease diagnosis.

## Key findings

- The multi-phase model achieved a test F1-score of 77.0%, outperforming single-phase models.
- Explainability techniques visualized heart regions relevant to disease classification.
- Deep learning models can effectively and automatically classify CVDs from cine MRI.

## Abstract

Cardiovascular diseases (CVDs) are the major cause of death worldwide. Magnetic resonance imaging (MRI) is the gold standard modality for CVD diagnosis because of its ability to distinguish different types of soft tissues without the use of ionizing radiation. Cine MRI allows us to see the contractile function of the heart, and it is a safe method for patients with chronic kidney diseases. The aim of this work was to develop a deep learning model for automated classification of common CVDs from cine MRI while providing the model explainability. We investigated single-phase models based on either the end-diastolic (ED) or end-systolic (ES) phase using seven baseline deep learning models including ResNet, DenseNet and VGG. We then developed a multi-phase model including both ED and ES phases to incorporate cardiac function for CVD classification. While the single-phase model for the ED and ES phases yielded the highest test F1-scores of 71.0% and 76.0% respectively, the multi-phase model achieved a test F1-score of 77.0%. To better understand the model performance, we used explainability to visualize regions of the heart that exhibit characteristics of each disease. Our work has demonstrated that deep learning models can automatically and effectively classify CVDs from cine MRI while justifying classification, thus building trust from the clinical community.