Fractal Analysis of Sensory Images and Signals: Methodological Limitations, Translational Barriers, and Artificial Intelligence Integration

This is the authors' abstract. We don't add key points for this paper.

Fractal analysis has become a frequently used technique in sensory physiology for the evaluation of signal complexity, with potential applications ranging from the analysis of retinal vascular architecture to the quantification of changes in electroencephalographic recordings. Features such as fractal dimension and lacunarity may be able to detect discrete changes in both one-dimensional and two-dimensional signals associated with numerous physiological processes and disease development. However, their interpretation and translational potential are constrained by various limitations. In this concise review, we critically explore current applications of fractal analysis across major sensory domains and focus on the challenges that limit reproducibility and biological interpretation. We particularly focus on the biological ambiguity of fractal dimension and related measures, the non-equivalence of commonly used fractal metrics, acquisition-related variability, small datasets, and insufficient technical validation. Finally, we discuss issues related to the integration of fractal metrics into contemporary machine learning models, including challenges related to generalizability, overfitting, and poor explainability.