# Exploring unsupervised learning techniques for early detection of myocardial ischemia in type 2 diabetes

## TL;DR

This study uses unsupervised learning to explore differences in myocardial ischemia among type 2 diabetes patients, revealing distinct clinical and imaging patterns.

## Contribution

The novel use of clustering techniques to uncover variability in myocardial ischemia among T2DM patients through SPECT imaging.

## Key findings

- Two distinct patient clusters were identified with differing clinical and lifestyle characteristics.
- Cluster 1 showed more severe SPECT features like hypokinesis and impaired systolic coordination.
- No significant differences were found in ischemic burden or cavity size between clusters.

## Abstract

Myocardial ischemia can result in severe cardiovascular complications. However, the impact of clinical factors on myocardial ischemia in individuals with T2DM remains unclear. we applied a clustering approach to identify the variability in myocardial ischemia evaluated through Single-Photon Emission Computed Tomography.

Retrospective statistics derived from 637 T2DM patients with myocardial ischemia who participated in SPECT imaging at our hospital between January 2022 and September 2024 were gathered. Ischemia areas, cavity size, wall motion,ventricular contraction, cardiac systolic coordination, End-diastolic Volume, End-systolic Volume; Left ventricular injection fraction were assessed and analyzed. Clustering analysis of medical data in unsupervised learning, involving the elbow method and silhouette coefficient(cluster 1: 262; cluster 2: 375);.

The Healthcare information between two groups differed in multiple respects (1) Cluster 1 had the had the older patient(63.23 ± 12.31), longer average duration of diabetes(10.27 ± 8.77), higher Glycated Hemoglobin(HbA1c) values(7.69 ± 1.76), the higher level of serum creatinine (115.42 ± 106.18µmol/L);and a higher proportion of patients with insulin treatment(40.5%) (2).Cluster 1 had more males(68.8%),higher proportion of patients with smoking history(44.5%), the higher level of Cholesterol(3.96 ± 1.12mmol/L),serum uric acid (406.78 ± 135.24µmol/L),Low-density lipoprotein cholesterol(2.08 ± 0.32mmol/L),and was more prone to statin therapy (6.1%).The SPECT features differed across the various clusters (1):Cluster 1 had higher proportion of Hypokinesis(38.2%),poor ventricular contraction(57.6%),Impaired Cardiac systolic coordination(63.7%),and abnormal LVEF(81.3%) (2).Cluster 2 had a higher proportion of total ischemia(11.5%) and abnormal ESV(52.8%) (3).There was no significant difference in Ischemia areas, Cavity size, Involved segments, and EDV.

Although the unsupervised clustering approach revealed differences in various clinical and imaging characteristics, no significant differences were observed in ischemic burden, cavity size, involved segments, or EDV.

Summary of the study design and key findings.

A composite image detailing a study on myocardial ischemia in Shanxi, China. The left panel describes a retrospective study of 637 T2DM patients with graphics indicating demographics, lifestyle, clinical, laboratory examination, and treatment. It includes a map of Shanxi province and a depiction of a Single-Photon Emission Computed Tomography (SPECT) scanner. The top right panel shows graphs for determining the optimal K using the Elbow method and Silhouette coefficient. The bottom right panel depicts findings from T-tests and chi-square tests, highlighting statistical significance with bar charts comparing clinical and SPECT characteristics.