HistoSuite: AI-powered histopathology

Following tissue collection and standard histological preparation, tissue cross-sections are digitised at high resolution. Whole slide imaging captures the complete tissue section at the magnification required for the analysis, producing image files that are stored, traceable and available for re-analysis without the degradation associated with physical slide storage. For studies that are performed elsewhere, the whole slide images are being transferred to our secured server.

HistoSuite applies trained deep learning models to each whole slide image. The models identify and delineate tissue compartments and lesion regions at pixel level, distinguishing between tissue types and pathological features with a precision that manual review cannot match at equivalent throughput.

The platform covers multiple organ systems including cardiovascular, liver, renal and CNS models, with model variants specific to the staining protocols and lesion morphologies relevant to each.

Analysis results are returned as structured quantitative data rather than categorical grades. Where manual scoring produces results such as a MASH activity score or an atherosclerosis grade, HistoSuite can expand these data as it produces also continuous measurements: plaque area in square micrometres, collagen fraction as a percentage of tissue area, or inflammatory cell density per region of interest. These outputs support parametric statistical analysis and provide greater sensitivity to detect treatment effects, particularly where effect sizes are moderate.

Quantification of aortic root lesion area, plaque composition (lipid core, fibrous cap, macrophage infiltration, smooth muscle cell content) and lesion progression over time are among the most demanding histopathological tasks in preclinical cardiovascular research.

Manual segmentation of HPS-stained whole slide images is time-consuming and operator-dependent. HistoSuite automates the full segmentation pipeline, from region of interest identification to plaque boundary delineation and compositional classification, delivering reproducible quantitative outputs in a fraction of the time.

In preclinical MASLD and MASH studies, histological scoring covers steatosis, ballooning, inflammation and fibrosis. HistoSuite quantifies collagen deposition using Sirius Red-stained sections, providing continuous measurements of fibrosis area rather than the ordinal F0 to F4 staging that pathologists apply.

TNO's diabetic/chronic kidney disease model develops glomerulosclerosis, tubulo-interstitial fibrosis and albuminuria in the context of obesity, dyslipidaemia and hypertension. Our kidney models analyzes all relevant kidney structures and provides quantifiable morphological features.

In CNS efficacy models, relevant histological endpoints include neuroinflammation markers, demyelination and cellular loss in defined anatomical structures. HistoSuite applies region-aware segmentation that accounts for the spatial heterogeneity of CNS tissue, enabling quantification within anatomically defined areas rather than across the slide as a whole.

HistoSuite operates across the full range of TNO's preclinical efficacy models. This includes the TNO Ldlr-/-.Leiden mouse for obesity, MASLD, MASH and liver fibrosis; the APOE*3-Leiden and APOE*3-Leiden.huCETP models for cardiovascular and lipid indications; the diet-induced diabetic/chronic kidney disease model ; and fibrosis models across skin, lung, liver and kidney.

HistoSuite outputs are integrated with the broader dataset generated in TNO efficacy studies, including gene expression, plasma biomarker panels, metabolic readouts and, where relevant, metabolic flux data from our AMS microtracer platform. The result is a multidimensional dataset in which tissue-level findings are interpreted alongside circulating and mechanistic readouts, providing a more complete picture of compound activity than any single modality alone.