[Paper Review] Cancer Detection via Determination of Fractal Cell Dimension

W. Bauer, Charles D. Mackenzie|ArXiv.org|Jul 6, 1995

Fractal and DNA sequence analysis3 references16 citations

TL;DR

This paper proposes a novel method to detect cancer by quantifying the fractal dimension of cell surface perimeters using electron microscopy and automated image analysis. By applying the box-counting method to digitized cell images, the authors demonstrate that hairy-cell leukemia lymphocytes exhibit significantly higher fractal dimensions (mean ~1.34) than healthy lymphocytes (all <1.28), enabling clear distinction between cancerous and non-cancerous cells with high accuracy.

ABSTRACT

We utilize the fractal dimension of the perimeter surface of cell sections as a new observable to characterize cells of different types. We propose that it is possible to distinguish cancerous from healthy cells with the aid of this new approach. As a first application we show that it is possible to perform this distinction between patients with hairy-cell lymphocytic leukemia and those with normal blood lymphocytes.

Motivation & Objective

• To develop a quantitative, objective method for distinguishing cancerous from healthy cells using fractal geometry.
• To investigate whether the fractal dimension of cell surface contours can serve as a reliable biomarker for cancer detection.
• To apply automated image processing techniques to extract and analyze fractal dimensions from electron micrographs of human blood cells.
• To evaluate the potential of fractal dimension as a diagnostic tool in clinical pathology, particularly for hematological malignancies.
• To explore the broader applicability of this method to other cancers, including breast cancer, and its potential for early detection.

Proposed method

• Digitizing electron micrographs of cell sections at high resolution (750×675 pixels) with 256 gray levels.
• Using grey-level histogram analysis to automatically set thresholds for cell membrane and nucleus, selecting a threshold of 80 for membrane detection.
• Applying a cluster-recognition algorithm to remove small speckles (below 300 pixels) that are artifacts, preserving the true cell surface.
• Performing partial derivatives in x and y directions to extract the surface as a black line on a white background, representing the cell boundary.
• Calculating the fractal dimension using the box-counting method over a range of box sizes (ξ from 1/8 to 1/64), fitting the log-log plot of N(ξ) vs. ξ to a power law.
• Using linear regression on the l-interval from 3 to 6 (ξ from 1/8 to 1/64) to estimate the fractal dimension d, with a correlation coefficient r > 0.999.

Experimental results

Research questions

• RQ1 Can the fractal dimension of a cell's surface perimeter reliably distinguish between healthy and cancerous lymphocytes?
• RQ2 Does the method of box-counting fractal dimension estimation yield consistent and reproducible results on biological cell images?
• RQ3 Is there a measurable difference in fractal dimension between healthy lymphocytes and those affected by hairy-cell leukemia?
• RQ4 Can automated image processing techniques effectively isolate and analyze cell surface contours without distorting their natural complexity?
• RQ5 Can this method be generalized to detect other types of cancer based on morphological surface complexity?

Key findings

• The fractal dimension of healthy lymphocytes was consistently below 1.28, with no cells exceeding this value.
• In contrast, a significant proportion of lymphocytes from hairy-cell leukemia patients had fractal dimensions above 1.28, with a mean of 1.34 for the sample studied.
• The box-counting method yielded a fractal dimension of d = 1.34 for a representative hairy-cell leukemia cell, with a correlation coefficient of r = 0.99975 for the power-law fit.
• The method demonstrated robustness to threshold variations within ±5 and cluster size variations from 100 to 500 pixels, showing minimal impact on final results.
• The technique successfully distinguished cancerous from non-cancerous cells based on surface complexity, with no overlap in fractal dimension values above 1.28 in the healthy group.
• The authors confirmed the method's accuracy by testing it on known fractals (Koch’s Snowflake, Sierpinsky’s Carpet), achieving results within ±0.03 of theoretical values.