Development and Validation of an Automated Image-Based Method for Prognostic Stratification in Lung Adenocarcinoma using Immunohistochemical Markers

We present an automated image-based framework for quantifying biomarker expression in immunohistochemical (IHC) images and introduce the Marker Expression Level (MEL) score as an objective alternative to visual H-scoring. MEL addresses key limitations of existing automated methods by integrating tissue-region segmentation with tumour-area–weighted marker quantification, normalization to mitigate inter-sample and inter-batch staining variability, and robust quantification of nuclear, cytoplasmic, and membrane-associated biomarkers. The complete pipeline was developed as an open-source Fiji/ImageJ tool and is publicly available.

We evaluated MEL using IHC images from 128 lung adenocarcinoma patients stained for BRCA1, GLUT1, and QKI, three markers forming a prognostic signature previously described by our group. Tissue regions were segmented using a machine learning-based Fiji plugin, and MEL scores were calculated within tumour regions. A normalization strategy accounted for inter-TMA staining variability. Agreement with visual H-scores was assessed using Spearman correlation. Prognostic performance was evaluated using Cox regression to derive a Combined Prognostic Index (CPI) incorporating MEL scores and tumour stage.

Segmentation achieved high accuracy across tissue regions (Dice coefficients 0.69–0.99), with lower performance in highly immune-infiltrated stromal areas. MEL scores significantly correlated with visual H-scores for all markers (ρ = 0.61–0.69, p < 0.001). Normalization improved agreement for GLUT1 (ρ = 0.39 to 0.62). The derived CPI stratified patients into distinct risk groups with significant differences in disease-free and overall survival (p < 0.05), comparable to visual scoring.

MEL provides a reproducible and scalable alternative to expert visual H-scoring and standard semi-automated approaches such as QuPath. Its integration with prognostic modelling reproduced a validated survival signature with performance comparable to expert scoring, supporting clinical and translational applicability. MEL is not a marginal refinement of existing digital scoring methods, but rather a clinically oriented and biologically adaptable framework designed to improve the robustness and reproducibility of automated IHC quantification.