Medical Image Texture: Biological Phenotype or Technical Fingerprint? From Early Texture Analysis to Radiomics and Clinical Decision-Making

Gotowe do wklejenia w pole Description. Edytor Zenodo obsługuje formatowanie, więc pogrubienia i akapity przeniosą się poprawnie. Preprint of a narrative review. Medical images have long been interpreted using qualitative descriptors of spatial organization such as homogeneity, heterogeneity, coarseness, trabeculation, or necrosis. Quantitative texture analysis emerged to translate these perceptual patterns into reproducible numerical descriptors and subsequently became a foundation of radiomics. However, image texture is not determined by tissue biology alone. It also reflects acquisition physics, spatial resolution, noise, reconstruction, preprocessing, normalization, discretization, and segmentation, creating a central ambiguity: a feature may represent a biological phenotype, a technical fingerprint, or both. This review traces the evolution of quantitative texture analysis from Haralick descriptors, COST B11, and MaZda to handcrafted radiomics, deep radiomics, and foundation representations. It examines the physical meaning and technical robustness of texture across MRI, CT, and projection radiography, with ultrasound speckle considered as a limiting case; asks why radiomics has generated far more publications than clinically deployed tools; and considers the transition from paper-grade to decision-grade evidence. The review proposes a Biological–Technical Texture Framework, in which a texture-derived measure becomes a credible imaging biomarker only when measurement validity, biological validity, and clinical validity converge. The central condition is formalized as a Biological–Technical Ratio, BTR = σ²_biological / σ²_technical. Because BTR = ICC / (1 − ICC), the construct is a transformation of the intraclass correlation coefficient, which allows conventional reliability categories to be read directly as interpretable variance ratios: an ICC of 0.75 corresponds to a biological signal only three times larger than the technical noise, whereas individual-patient decisions require ICC ≥ 0.90, that is BTR ≥ 9. The future of quantitative imaging may therefore depend less on extracting increasingly large feature sets than on identifying a smaller number of technically robust, biologically anchored, externally validated, and clinically actionable imaging biomarkers. Contents. Sixteen sections, eight figures, and seven tables, covering the history of texture analysis; the dual biological and technical origin of image texture; modality-specific behaviour in MRI, CT, radiography, and ultrasound; deep-learning reconstruction as a new source of feature variability; segmentation variability; the reproducibility crisis and the standardization response (IBSI, CLEAR, METRICS, RQS 2.0); handcrafted versus deep and foundation representations; the proposed framework; and the regulatory and reporting pathway under the EU Medical Device Regulation and the EU AI Act. The reference list comprises 96 sources. Ethics approval and consent to participate. This narrative review is based exclusively on previously published literature and publicly accessible scientific information. No new individual-level or identifiable patient data were collected or analysed, no human or animal participants were recruited, and no biological material was used. Institutional ethics committee approval and informed consent were therefore not required. Data availability. No new datasets were generated or analysed. All information supporting the discussion is derived from the published sources cited in the manuscript. Funding. This research received no external funding. Competing interests. The authors declare no competing interests relevant to this work. Use of generative AI and AI-assisted technologies. During preparation of this manuscript the authors used AI-assisted tools for language editing, structural organization, literature-search support, and refinement of figure concepts, and AI image-generation tools to produce schematic figures. All AI-assisted text, bibliographic details, and figures were critically reviewed and verified by the authors, who take full responsibility for the content of the final manuscript. Note on Figure 3. Figure 3 is a simulated example. The values shown are illustrative and do not represent measurements from a specific scanner or phantom. Corresponding author. Michał Strzelecki, [email protected]