Information Entropy-Based Framework for Quantifying Tortuosity in Meibomian Gland Uneven Atrophy

Kesheng Wang, Xiaoyu Chen, Chunlei He, Fenfen Li, Xinxin Yu, Dexing Kong, Shoujun Huang, Qi DaiarXiv 2025 · DOI · arXiv

Introduction

In clinical diagnostics, the "tortuosity" or irregular curvature of anatomical structures often serves as a critical biomarker for disease progression. For instance, the abnormal winding of retinal vessels can indicate retinopathy. However, traditional quantification methods, such as the arc-chord ratio, frequently rely on comparisons to idealized straight lines. This approach is fundamentally flawed for biological structures like meibomian glands, which possess natural physiological curves. Comparing these naturally curved structures to straight lines introduces significant "noise," obscuring pathological changes like the uneven atrophy caused by Demodex mite infestations.

Methodology: The Information Entropy-Based Framework (IEBF)

To address the limitations of linear baselines, researchers have developed the Information Entropy-Based Framework (IEBF). This system shifts the focus from absolute geometric length to the "disorder" of a curve relative to a biologically plausible reference.

The IEBF operates in three distinct stages:

  1. Boundary Extraction: Deep learning models segment the glands, followed by an Active Contour Model ("Snakes") to refine edges into smooth, continuous curves.
  2. Entropy-Based Quantification: Instead of a straight line, the framework utilizes a non-linear reference curve that mimics healthy anatomy. It calculates point-to-point distance differences (Δy) between the target and reference curves. These differences are mapped to a probability distribution to calculate Information Entropy. High entropy signifies high structural disorder, indicating pathological tortuosity.
  3. Frequency Correction: Fourier Transforms decompose curves into low-frequency (structural shape) and high-frequency (noise) components, allowing the system to filter out imaging artifacts while preserving diagnostic data.
Figure 3: Conceptual illustration of the IEBF, showing the calculation of distance differences (Δy) between a target curve (f1) and a reference curve (f2) to determine entropy.

Experimental Validation

The framework was validated using both synthetic and clinical data. Numerical simulations involving 5,000 curves with varying Gaussian noise levels confirmed the algorithm's stability. Subsequently, a clinical study was conducted with 105 subjects (35 Demodex-negative and 70 Demodex-positive) from the Wenzhou Medical University Eye Hospital. Meibography images were processed to extract gland boundaries for entropy analysis.

Key Results

The IEBF demonstrated superior diagnostic performance compared to traditional geometric metrics. The framework achieved an Area Under the Curve (AUC) of 0.858, significantly outperforming the chord-to-arc ratio (AUC 0.811) and total variation methods (AUC 0.656).

Statistical analysis revealed a highly significant difference ($p < 0.001$) in entropy scores between the infected and healthy groups. With a sensitivity of 0.786 and a specificity of 0.857, the IEBF provides a robust, objective metric for identifying uneven atrophy, a hallmark of Demodex-induced Meibomian Gland Dysfunction (MGD).

Figure 6: Boxplot comparison of entropy-based tortuosity values between Positive and Negative Demodex groups, showing a clear statistical separation.

Discussion and Future Directions

The core innovation of the IEBF lies in rejecting the "straight line fallacy." By acknowledging natural biological curvature, the system becomes sensitive to true pathological warping. However, the framework's accuracy is contingent upon the selection of an appropriate reference curve. Future work will focus on leveraging historical patient data to generate personalized reference baselines and integrating multimodal imaging to further enhance diagnostic precision.

Summary updated

BibTeX
@misc{wang2025information,
  author = {Kesheng Wang and Xiaoyu Chen and Chunlei He and Fenfen Li and Xinxin Yu and Dexing Kong and Shoujun Huang and Qi Dai},
  title = {{Information Entropy-Based Framework for Quantifying Tortuosity in Meibomian Gland Uneven Atrophy}},
  year = {2025},
  eprint = {2507.18135},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2507.18135},
}
Figure