ROI-based Grad-CAM intensity on chest radiographs was significantly higher in females than in males in the 0-12 months preceding heart failure onset in peritoneal dialysis patients.
Cohort (n=73)
No
Do longitudinal chest radiograph signatures analyzed by AI differ between males and females preceding heart failure onset in peritoneal dialysis patients?
AI-based analysis of longitudinal chest radiographs suggests there may be sex-specific differences in radiographic patterns in the 12 months preceding clinical heart failure onset in peritoneal dialysis patients.
Absolute Event Rate: 0.3% vs 0.07%
p-value: p=0.045
Abstract Background Heart failure (HF) is a leading cause of morbidity and mortality among individuals with kidney failure with replacement therapy (KFRT). Although biological sex influences cardiovascular disease, it remains unclear whether and how longitudinal chest radiograph (CXR) patterns before HF onset differ between males and females in high-risk KFRT populations. Methods We conducted a retrospective single-center cohort study of adults who initiated peritoneal dialysis over a 20-year period and were classified as HF ( n = 39; 30 male, 9 female) or non-HF ( n = 34; 21 male, 13 female). Longitudinal CXRs were analyzed using a convolutional neural network with gradient-weighted class activation mapping (Grad-CAM). A cardiac region of interest (ROI), introduced after exploratory Grad-CAM visualization suggested predominant attention in the cardiac region, was applied to all CXRs. HF onset was defined as time 0. CXRs were grouped into 0–12, 12–24, and 24–36 months before HF onset, and ROI-based Grad-CAM intensity values were averaged within each interval for each patient. Patient-level mean ROI-based Grad-CAM intensity values were then compared between males and females within each interval. Results Baseline clinical characteristics were similar between HF and non-HF groups. In the time-aligned patient-level analysis, ROI-based Grad-CAM intensity was higher in females than in males in the 0–12-month interval before HF onset (median 0.30 versus 0.07, p = 0.045), whereas no significant between-sex differences were observed in the 12–24-month ( p = 0.558) or 24–36-month ( p = 0.143) intervals. Conclusions Routine longitudinal CXRs in peritoneal dialysis patients may contain differences in radiographic attention patterns between males and females preceding clinically recognized HF. However, these findings are exploratory and hypothesis-generating and require validation in larger, multicenter studies.
Muto et al. (2026) conducted a cohort in Heart failure in peritoneal dialysis patients (n=73). Female sex vs. Male sex was evaluated on Patient-level mean ROI-based Grad-CAM intensity 0-12 months before HF onset (p=0.045). ROI-based Grad-CAM intensity on chest radiographs was significantly higher in females than in males in the 0-12 months preceding heart failure onset in peritoneal dialysis patients.