Ensuring reliable digital pathology: a comparative analysis of HistoQC and PathProfiler for artefacts detection in...

Whole-slide images (WSIs) offer high-resolution views of tissue but are often compromised by artefacts that hinder human and AI interpretation, potentially leading to diagnostic errors. Open-source tools like HistoQC and PathProfiler aim to assess image quality, but their comparative performance in prostate cancer WSIs remains unexplored.

We evaluated HistoQC and PathProfiler using 240 WSIs from the TCGA-PRAD dataset: 120 artefact-free slides and 40 for each of three artefact categories-usability, out of focus and staining. A three-phase approach was used: (1) PathProfiler scores were normalized and validated via ROC analysis; (2) HistoQC scores were derived using regression models (SVR, XGBoost, LR, LightGBM, GP), trained on 192 slides with 5-fold cross-validation; (3) Tools were compared using Spearman's correlation, MAE, MSE, MAPE, and Wilcoxon-Mann-Whitney tests (p < 0.05). Performance was evaluated on a 48-slide test set.

PathProfiler showed strong artefact detection (AUCs: 0.960 staining, 0.948 out of focus, 0.921 usability). HistoQC, enhanced by regression models (SVR and LightGBM), demonstrated significant correlation with PathProfiler (ρ = 0.790-0.672-0.537, all p < 0.001) and comparable AUCs (0.842-0.858-0.826). Both tools effectively distinguished between clean and artefact-laden slides (p < 0.05) across all categories.

HistoQC and PathProfiler reliably assess WSI quality in prostate cancer. PathProfiler offers efficiency for clinical use, while HistoQC provides adaptable scoring via machine learning. Together, they can enhance diagnostic accuracy and support integration of AI in digital pathology workflows.