Pathological classification of non-ischaemic dilated cardiomyopathy based on deep learning

Area of Science:

• Cardiology
• Computational Pathology
• Precision Medicine

Background:

• Non-ischaemic dilated cardiomyopathy (NIDCM) presents heterogeneous clinical phenotypes and disease progression, lacking precision diagnostics and treatments.
• Heart failure (HF) and heart transplantation (HTx) are common outcomes for NIDCM patients.
• Identifying high-risk NIDCM patients for malignant arrhythmia (MA) and rapid progression is crucial.

Purpose of the Study:

• To stratify NIDCM patients based on pathological features using deep learning computational pathology (DL-CPath).
• To identify high-risk NIDCM subgroups associated with malignant arrhythmia (MA) and rapid progression to end-stage HF.
• To correlate pathological subgroups with clinical phenotypes and outcomes.

Main Methods:

• Analysis of 3516 heart tissue slides from 293 NIDCM-HTx patients using DL-CPath.
• Unsupervised clustering to define pathological subgroups (PGs): PGA, PGB, and PGC.
• Correlation of PGs with clinical data, including MA rates, time to HTx, and blood biomarkers.

Main Results:

• Three distinct pathological subgroups (PGA, PGB, PGC) were identified.
• PGA exhibited interstitial fibrosis, cardiomyocyte vacuolization, and myocyte disarray, associated with the highest MA rates and shortest diagnosis-to-HTx interval.
• PGA patients had elevated injury biomarkers; PGB showed extensive fibrosis and reduced ejection fraction; PGC had mildest alterations.

Conclusions:

• DL-based pathological classification effectively identified clinically meaningful imaging features in NIDCM.
• Distinct pathological subgroups (PGs) demonstrate unique histopathological and clinical characteristics, enabling risk stratification.
• This approach highlights potential for precision medicine strategies in NIDCM management.