PathQC: Determining Molecular and Structural Integrity of Tissues from Histopathological Slides

Area of Science:

• Computational pathology
• Biobanking
• Artificial intelligence in medicine

Background:

• Assessing tissue integrity is crucial for biobanks but current methods are destructive or manual and not scalable.
• Existing quality control (QC) methods focus on imaging, not sample integrity, limiting biobank utility.
• There's a need for non-destructive, scalable methods to evaluate molecular and structural integrity of biospecimens.

Purpose of the Study:

• To develop and validate PathQC, a deep learning framework for predicting tissue RNA Integrity Number (RIN) and autolysis from H&E stained whole-slide images.
• To provide a scalable, non-destructive method for sample-quality control in biobanks.
• To enable retrospective analysis of tissue integrity across large cohorts.

Main Methods:

• PathQC utilizes a digital pathology foundation model (UNI) to extract morphological features from H&E slides.
• A supervised model predicts RNA Integrity Number (RIN) and autolysis scores based on extracted morphological features.
• The framework was trained and applied to the Genotype-Tissue Expression (GTEx) cohort (25,306 samples, 29 tissues).

Main Results:

• PathQC achieved an average Pearson correlation of 0.47 for RIN and 0.45 for autolysis across tissues.
• High performance was observed in specific tissues: adrenal gland (RIN R=0.82) and colon (autolysis R=0.83).
• A pan-tissue model was developed, capable of predicting RIN and autolysis scores for various tissue types.

Conclusions:

• PathQC offers a scalable solution for assessing tissue molecular and structural integrity using routine H&E images.
• This deep learning approach enhances biobank quality control and facilitates retrospective studies on tissue quality.
• The framework supports quality assessment across diverse tissues and collection sites, improving data reliability.