Accurate pathologic classification of aggressive myocarditis improves insights into patient precise medicine

Fig. 1. Pipeline of prognostically relevant myocarditis classification based on pathologic WSIs. (a) Overview of myocarditis classification through efficient pathologic image feature extraction and prediction of prognostic risk factors. The integrated patient-level features will be used to calculate risk values using the Cox model. Subsequently, by ranking these risk values from low to high, all patients can be categorized into three distinct groups. (b) Elaboration on prognostic risk prediction and exploration of pathologic appearances related to subtypes. Taking patient i containing n WSIs as an example, our process initiates by partitioning each WSI into patches. Subsequently, we extract patch-level features utilizing our foundation model. WSI-level features with a size of 1 × 512 are subsequently acquired through a feature aggregation network. The identification of predicted regions of interest involves mapping through a deep attention mechanism, aiding in the exploration of subclasses expressing relevant pathological features.

Fig. 2. Heart transplantation free survival stratified by phenogroup. (a) K-M survival curve of heart transplantation in different pathological groups. (b) K-M survival curve of heart transplantation in different genders. (c) K-M survival curve of heart transplantation with or without proviral infection. (d) K-M survival curve of heart transplantation with or without diuretics. (e) Risk forest map of clinical prognostic related indicators by Cox analysis.

Fig. 3. Extraction of extremal tiles associated with prognosis to identify regions of interest. (a) Attention scores were extracted from a total of 498 cases of whole slide images (WSI), with patients displaying fast transplant progression at the bottom and those with slow transplant progression at the top. (b) Extreme tiles related to overall fast and slow transplant progression in all patients with myocarditis were extracted and reviewed by pathologists. (c) Histological analysis revealed that tiles with high attention scores and rapid graft progression exhibited inflammatory cell aggregation and giant cell mass, while tiles with low attention scores and slow graft progression showed fibrosis, vacuolation, and disordered heart arrangement.

Fig. 4. Genetic mutation oncoplot and differential clinical indicators of three phenogroups were enriched. (a) Displays of the proportion of mutations present in each of the three PG groups. (b) A oncoplot map that visualizes all P/LP/VUS mutations. Additionally, it includes a heat map depicting clinical characteristics that exhibit significant differences among the three groups.

Fig. 5. Transcriptomic profiling and biomarker validation in aggressive myocarditis phenogroups. (a) Overview for transcriptome analysis. (b) Venn diagram of differentially expressed genes (DEGs) from pairwise comparisons between pathological phenogroups (PG1, PG2, PG3). (c) GSEA enrichment analysis of DEGs for PG3 vs PG1 (top) and PG3 vs PG2 (bottom). (d) The ssGSEA immunoinfiltration heatmap derived from the transcriptome data of different PGs. (e) Representative immunohistochemical staining for CD3 (T lymphocytes), CD68 (macrophages), and MPO (neutrophils) in myocardial tissues. (f) Quantitative histomorphometry of IHC staining intensity (*P < 0.05, ⁎⁎⁎P < 0.001). (g) Forest plot of hazard ratios (HR) for 14 immune-related genes significantly associated with heart transplantation-free survival (P < 0.05), stratified by median gene expression. (h) K-M curves for IL6ST (highest HR) and IFNA5 (lowest HR). (i) ROC analysis of serum IL6ST and IFNA5 protein levels in an independent validation cohort. (j) Based on the optimal threshold obtained from ELISA (IL6ST/IFNA5), a KM survival analysis was conducted to examine the recurrence of symptoms.

Fig. 6. Clinical significance of phenotypic grouping and construction of decision tree model. (a) Naming of clinical features of the three groups of pathological phenotypes; (b) The coefficient ranking of the three pathological groups and their related clinical features; (c) Decision tree model construction.