Complex shape markers can detect alterations in the spatial distribution of cell nuclei in human lung squamous cell carcinoma: a useful tool for automatic analysis?

doi:10.21203/rs.3.rs-4694025/v1

Download PDF

Research Article

Complex shape markers can detect alterations in the spatial distribution of cell nuclei in human lung squamous cell carcinoma: a useful tool for automatic analysis?

https://doi.org/10.21203/rs.3.rs-4694025/v1

This work is licensed under a CC BY 4.0 License

You are reading this latest preprint version

Lung cancer is the leading cause of cancer deaths. The use of computational methods to quantify changes not perceptible to the human eye is growing in digital pathology imaging, improving detection rates quickly and at low cost. Therefore, the present study aims to use computational complex shape markers as tools for automated analysis of the spatial distribution of cells in microscopic images of squamous cell lung carcinoma (SqCC). Photomicrographs from pathology glass slides in the database LC25000 were used. The fractal dimension and lacunarity of the lung cell nuclei statistically changed in SqCC compared to the control. The multifractal analysis showed a significant difference in Dq, α, and f(α) for all values of q (-10 to + 10), with a greater increase for more positive q values. The number of cells, circularity, area, and perimeter also changed in SqCC images. However, the parameters aspect ratio, roundness, and solidity did not show statistical differences between the SqCC and benign tissue. The complex shape markers with the greatest changes in this study were the f(α) value in multifractality (53%) and lacunarity (41%). In conclusion, the automated quantification of the spatial distribution of cell nuclei can be a fast, low-cost tool for evaluating the microscopic characteristics of SqCC; therefore, the complex shape markers could be useful methods for software and artificial intelligence to detect lung carcinoma.

Lung cancer

fractal

lacunarity

multifractality

morphometry

automation

Lung cancer is the leading cause of cancer-related death globally (Adams et al. 2023). Lung cancer has an estimated 2.2 million new cases and 1.8 million deaths in 2020. Its incidence and mortality rates differ around the world, reflecting varying patterns of tobacco smoking, genetics, exposure to environmental risks (such as biomass fuels, arsenic, radon, industrial carcinogens, and air pollution), hereditary cancer syndromes, and previous chronic lung diseases (Leiter et al. 2023; WHO 2023).

Lung cancer can be broadly divided into small cell lung carcinoma (SCLC) and non-small cell lung carcinoma (NSCLC). The most common histological types of NSCLC are adenocarcinoma (40%) and squamous cell carcinoma (SqCC) (25%) (Leiter et al. 2023). Pathologists commonly need to distinguish SCLC from NSCLC because of the several different treatments; similarly, NSCLC subtypes also need to be identified due to a shift in epidemiology and an increasing number of target-specific therapies. Lung SqCC microscopically shows a malignant epithelial tumor with keratinisation and/or intercellular bridges that arise from bronchial epithelium (Paech et al. 2011). Adenocarcinoma and SqCC can be identified by thyroid transcription factor 1 (TTF1) and p40 staining, even in the case of poorly differentiated NSCLC with small biopsy and cytologic specimens. However, as with any technique, there are adverse factors in the selection of the antibody panel, antibody clones, and interpretation of the staining (Yatabe et al. 2019).

Currently, molecular markers and high-cost tests detect SqCC; however, these expensive and time-consuming tests limit access for low-income populations, time-dependent patients, and hinder the implementation of broad-scale monitoring programs. Continued development of computational methods is particularly important if lung cancer screening is to be implemented more broadly, as well as the need for standardized image quality and lung nodule evaluation (Adams et al. 2023). The fields of computational methods and artificial intelligence (AI) are rapidly growing and have the potential to affect many aspects of our lives, including the healthcare field, especially its potential utility for diagnosing disease (Borkowski et al. 2019). AI algorithms have demonstrated high sensitivity (83–97%) and accuracy (82–98%) for lung nodule detection, particularly for smaller nodules, reducing the false-positive rate. It is worth mentioning that the software generally requires a human operator to review all nodule candidates identified to determine whether they are a true nodule. Thus, the use of AI focuses on playing a role as a first or second reader to enhance the efficiency of lung cancer detection (Adams et al. 2023).

Digital pathology is already used in many laboratories around the world, where microscopes are linked to digital cameras, scanners, computers, and monitors, opening new horizons for histopathology. In this context, computers can assist in the detection of pathological lesions by scanning the image, quantifying the characteristics of the lesion, and improving the information and classification associated with diseases (Melo et al. 2022). Many detection methods are expensive and difficult to implement in healthcare systems because they consume a lot of time, skilled labor, expensive equipment, and materials. On the other hand, using computer-based methods to detect changes has proven to be a low-cost, efficient, and widely distributed tool around the world. In addition, computer methods have practical applications for pathologists to detect changes not perceptible to human eyes (Melo et al. 2022; Tenorio et al. 2017).

The World Health Organization has called for worldwide efforts to identify “best buys” and other cost-effective priority strategies for cancer prevention and control by developing precise standards and tools to guide the planning and implementation of interventions (Barbora et al. 2023). Several new technologies are emerging to improve the efficiency of lung cancer detection. Computational methods and AI aiming to assess the malignancy probability could help in classifying lung nodules as benign or malignant. These tools can optimize nodule management by reducing unnecessary work-up of benign nodules, decreasing time to diagnosis of malignant nodules, and minimizing inter-reader variability. Furthermore, novel approaches combining algorithms could reduce the false-positive rate in lung cancer (Adams et al. 2023; Melo et al. 2022).

The lung follows a branching fractal pattern where the various alveoli and bronchioles exhibit self-similarity across different scales. Fractal geometry methods, unlike Euclidean methods, are able to quantify complex characteristics of structures through fractal dimension and lacunarity measurements. Therefore, fractal methodology has been used to quantify the complex physiology of the lung at various branching levels (Mambetsariev et al. 2019). According to Lennon et al. (2015), fractal geometry can be used effectively to estimate the geometrical complexity and irregularity of shapes and patterns. Thus, the present study aims to develop a new perspective in the microscopic analysis of SqCC lung cancer using complex shape markers as tools for automated detection methods.

2.1 EXPERIMENTAL DESIGN

Photomicrographs (768 x 768 pixels) from pathology glass slides were captured by a Leica MC190 HD camera (Wetzlar, Germany) connected to an Olympus BX41 microscope (Center Valley, PA), as previously described by Borkowski et al. (2019A). The images are available in the database histopathological images of lung and colon cancer LC25000, which contains a set of histopathological images with the classes: I) Benign lung tissue (Control); and II) Squamous cell carcinoma of the lung (SqCC). All images in LC25000 are de-identified, HIPAA compliant, validated, and freely available for AI researchers (Borkowski et al. 2019B). 200 photomicrographs of benign lung tissue and 200 photomicrographs of SqCC were processed using ImageJ software (National Institutes of Health - NIH, USA), with a macro made in our laboratory to obtain the thresholds (binarized images), as following:

rename("lung"); run("Colour Deconvolution", "vectors = H&E"); selectWindow("lung-(Colour_3)"); close(); close(); selectWindow("Colour Deconvolution"); close(); selectWindow("lung"); close(); selectWindow("lung-(Colour_1)"); run("8-bit"); run("Auto Threshold", "method = Li ignore_white white"); run("Median...", "radius = 3"); run("Invert LUT"); run("Fill Holes"); setOption("BlackBackground", true); run("Convert to Mask"); run("Analyze Particles...", "size = 150-Infinity show = Masks display clear"); selectWindow("lung-(Colour_1)"); close(); selectWindow("Mask of lung-(Colour_1)");

The binarized images were analyzed using the ImageJ plug-in FracLac (A. Karperien, Charles Sturt University, Australia), based on Melo et al. (2022).

2.2 FRACTAL DIMENSION

Images of spatial distribution of cell nucleus were used to calculate the fractal dimension by the box counting (Dbox) method. Briefly, the fractal dimension is calculated by covering the image with N(r) boxes, where N is the number of boxes and r is the length of one side of the box containing at least one point of the binarized cell nuclei. This was repeated with boxes of different sizes and plotted on a double log graph of N(r) as a function of r. The slope of this relationship with the inverted signal is the fractal dimension (box counting) and Ɛ is the smallest variation in the size of the boxes. This can be formally described by formula 1:

$$\:{D}_{bc}=-\underset{\epsilon\:\to\:0}{\text{lim}}\left[\frac{\text{log}N\left(r+\epsilon\:\right)-\text{log}N\left(r\right)}{\text{log}\left(r+\epsilon\:\right)-\text{log}r}\right]$$

(1)

2.3 LACUNARITY

The distribution of gaps in the image of the cell nuclei in the lung was described by lacunarity values. The image is covered with a series of grids, each grid containing a number of boxes in different sizes (ε) and orientations (g). The lacunarity value (Λ) is formally described by the formula 2:

$$\:\varLambda\:=\left[\sum\:_{g}\sum\:_{i}\left(1+{\left(\sigma\:|\mu\:\right)}^{2}\right)\right]/n$$

(2)

σ: standard deviation; µ: mean pixel value per box at a side size ε; n: number of box sizes in an orientation g.

2.4 MULTIFRACTALITY

In multifractal analysis of lung cells, the generalized dimension spectrum (Dq) is dependent on the variable q, an exponent that expresses fractal properties in q different scales for the image of lung. The value of q ranges from − 10 to + 10. The method describes fractal properties at different scales for the image obtained through the graph f(α) and α, where f(α) is a parameter related to the multifractal dimension and α is related to different scales. The graphs show the presence of high and low fractal exponents. These procedures can be formally described by the formulas 3, 4 and 5:

$$\:{D}_{q}=\underset{\epsilon\:\to\:0}{\text{lim}}\frac{1}{(q-1)}\frac{ln\:Z\:(q,\epsilon\:)}{ln\:\epsilon\:}$$

(3)

$$\:Z(q,\epsilon\:)=\sum\:_{i=1}^{{N}_{boxes}\:\left(\epsilon\:\right)}{\mu\:}_{i}{\left(\epsilon\:\right)}^{q}$$

(4)

Where, Nboxes describes the number of ε size boxes needed to cover the image of the structure.

$$\:{\mu\:}_{i}\left(\epsilon\:\right)=\frac{1}{T}{\sum\:}_{j=1}^{{N}_{{box}_{i}}}{\tau\:}_{j}$$

(5)

Where, µi is the pixel probability; τj is the number of pixels within the jth box; T: number of pixels for the whole image; ∑_iµ_i(ε) = 1; D_q = τ(q)/(q-1); α(q) = dτ(q)/dq; f(α(q)) = qα(q)-τ(q).

2.5 QUANTITATIVE MORPHOMETRY

Binarized images of the lung cells were used to calculate the following parameters: AREA: amount of two-dimensional space contained in the image (µm²); PERIMETER: outer boundary length of the image (µm); ROUNDNESS: 4 x [area] / π [major axis]², the inverse of aspect ratio; CIRCULARITY: 4 π x [area] / [perimeter]², where the value 1 indicates a perfect circle and the closer to 0, the more elongated the shape; ASPECT RATIO: major axis/minor axis; SOLIDITY: area/convex area.

2.6 STATISTICAL ANALYSIS

The results are expressed as mean (±) standard deviation. Statistical analysis was performed using the non-parametric test Mann-Whitney (p < 0.05).

3.1 HISTOPATHOLOGICAL ANALYSIS

No significant pathological alterations were observed on benign lung images (control). Microscopic images of the control lung showed cells with normal appearance, shape, size and staining. Pneumocytes 1 and 2, as well as connective tissue cells in the interalveolar septum were preserved. The pulmonary alveoli and vascularization were also normal. Alveolar macrophages were also observed in benign lungs, but in normal quantity and appearance (Fig. 1A). On the other hand, the images of SqCC lung cancer showed squamous epithelial differentiation (Fig. 1B). SqCC images showed atypical epithelioid cells, cytoplasmic eosinophilia, nests of tumor cells, areas of necrosis, keratinizing cells, keratin pearls and intercellular bridging. These results corroborate the findings described in the database LC25000.

3.2 FRACTAL DIMENSION AND LACUNARITY

The fractal dimension of the lung cell nuclei statistically increased (28%) in SqCC compared to control, p < 0.0001 (Fig. 2A). Lacunarity analysis showed a statistical reduction (41%) in the SqCC images, p < 0.0001 (Fig. 2B).

3.3 MULTIFRACTALITY

The multifractal analysis was performed using the generalized dimension values Dq versus q, the values f(α) and α (Fig. 3). All values of Dq, α and f(α) showed a statistical increase (p < 0.05) for all values of q (-10 to + 10) in SqCC images compared to benign tissue. Note the greater difference between the values of Dq and α for more positive values of q, from q0 to q + 10. The values at q + 10 increased by 34% for Dq, 36% for α, and 53% for f(α) in the SqCC images.

3.4 QUANTITATIVE MORPHOMETRY

Lung cell nuclei were also used to calculate the morphological parameters shown in Table 1. The aspect ratio, roundness, and solidity did not show statistical differences between the images of SqCC and benign tissue. However, the number of cells increased by 132% in SqCC compared to control. The total area and perimeter of cell nuclei increased by 95% and 42% in SqCC, respectively. Furthermore, the circularity of the lung cell nuclei was reduced by 7% in the SqCC images.

Table 1

Morphometric markers applied to the spatial distribution of the cell nucleus in human benign lung tissue (control, n = 200) and lung squamous cell carcinoma (SqCC) using the Image Dataset LC25000 (n = 200). The results were expressed by mean ± standard deviation.
	Control	SqCC	Z	p
Cell count	36.75 ± 11.27	85.33 ± 35.69	14.72	< 0.0001^*
Area	1036.18 ± 397.01	2022.91 ± 990.64	12.71	< 0.0001^*
Perimeter	133161.82 ± 27066.03	190924.70 ± 51280.40	12.51	< 0.0001^*
Circularity	0.70 ± 0.05	0.65 ± 0.06	8.37	< 0.0001^*
Aspect ratio	2020.72 ± 276.11	1970.49 ± 279.70	1.95	0.0603
Roundness	0.57 ± 0.05	0.58 ± 0.06	1.04	0.2969
Solidity	0.91 ± 0.02	0.88 ± 0.02	6.562	0.0601

* indicate a statistically significant difference between the experimental groups (p < 0.05).

According to the World Health Organization, lung cancer is the leading cause of cancer-related deaths worldwide, accounting for the highest mortality rates among both men and women. Lung cancer is a significant public health concern; however, this cancer is often diagnosed at advanced stages when treatment options are limited (WHO 2023). The use of immunohistochemistry has improved diagnostic accuracy in the classification of lung carcinoma, but the interpretation of the results remains challenging in some instances and the pathologists must be aware of many interpretation pitfalls (Yatabe et al. 2019). Often, molecular methods are difficult to apply to diagnosis due to the high costs and the long time required to complete the analysis (Mincione et al. 2015). In the same way, various equipment can have many disadvantages due to high cost, bulky and expert dependent subjective evaluations (Pala et al. 2022). Therefore, new technologies are emerging to try to improve the lung cancer detection at low cost, such as computational methods and AI applied to microscopic images of cancer.

Human eyes are very efficient at detecting patterns in images. However, the eye has limitations in its ability to analyze certain patterns of shapes, colors and sizes. In these cases, computerized image analysis can quantify changes in the structure that are imperceptible to human eyes. Furthermore, computer image analysis can reduce the time and cost of detections, with practical potential for use and distribution in healthcare systems (Melo et al. 2022; Tenorio et al. 2017).

According to da Silva et al. (2021), the identification of parameters that could successfully differentiate cancer images and aid in the diagnosis is of great interest. The present study aimed to find a different path in the automated analysis of microscopic images of human lung cancer by suggesting mathematical algorithms to quantitatively differentiate the spatial distribution patterns of the cell nuclei in SqCC and benign lung tissue. It is worth mentioning that the shape patterns shown by the spatial distribution of lung cells vary greatly and are difficult to differentiate quantitatively by human eye.

Overall, automated cancer detection involves, at least: 1) image preprocessing; 2) extracting identifiable features; and 3) correlating the features to the diagnosis (da Silva et al. 2021). A major difficulty in automating image analysis is segmentation, which is the separation of the analyzed structure and the image background, commonly resulting in black and white images (binarization). The present study used the spatial distribution of cell nuclei. The nuclei (region of interest) in routine staining present good contrast from the rest of the image (background), as they appear with color numbering in pixels clearly distinguishable from other color numbers in the background by the computer. Thus, in computer vision, the nucleus represents a good structure for image segmentation and binarization, and consequently, the results were satisfactory in the segmentation for automated nucleus analysis described in the methodology. Many previous studies used manual image processing to obtain results in microscopic images of cancer; therefore, these manual analyses are dependent on a human operator (subjective), reducing their automation potential. On the other hand, a highlight of the present study is the automation in the processing of lung microscopic images, which allows practical use in automatic software and AI.

Diagnosing cancer in small nodules and small biopsy samples is still a challenge; therefore, methods for analyzing small areas are of particular importance. In this sense, another highlight in our study is analyzing cell nucleus at high magnification in an area of a few micrometers, which allows the use of small lung samples with less invasive techniques and analysis of very small nodules.

The nucleus of lung cells varies greatly and form irregular/complex images even in benign tissue. However, the distribution and shape of nuclei in cancer can result in an even more complex image. Therefore, the present study used this increase in the complexity of the cell nucleus as shape marker to characterize lung carcinoma. Fractal dimension, lacunarity and multifractality are algorithms used to analyze the complexity/irregularity of biological images (Casey et al. 2024; Lee et al. 2014; Costa and Nogueira 2015; Sánchez-Chávez et al. 2024). According to da Silva et al. (2021), the formation of a tumor generally progresses with distortion of the tissue architecture, altering its complexity compared to the healthy counterpart.

Many structures in the human body, such as the distribution of cells in the lung, cannot be well described using classical Euclidean geometry, which considers lines, circles, squares, triangles and three-dimensional spaces, such as spheres, cubes and cylinders. The fractal dimension can measure how full the image space is occupied. As the fractal dimension has fractional values, fractal methods can better describe the degree of complexity and irregularity in images. Briefly applied to digital images, the fractal dimension can quantify how much of the image the structures occupy between dimensions; for example, between dimension 2 of a square (height and width) and dimension 3 of a cube (height, width and depth), and thus quantify fractional values to better describe complex biological structures. According to Pala et al. (2022), the larger the value of the fractal dimension, the more developed the properties of the structure at different scales. Therefore, the fractal dimension provides a relationship between fractal continuity and information richness (depth and detail), so if the detail and richness of the image are high, the fractal dimension value is also high. Many anatomical structures have been shown to demonstrate self-similarity properties, such as the lung with its bronchioalveolar ramifications. In this sense, fractal morphometry assessing architectural distortion and correlating with a specific pattern can enhance the recognition of histological structures (Mancini et al. 2023).

The present study demonstrated the ability of fractal dimension to detect significant differences between benign lung tissue and SqCC; showing an increase in the complexity/irregularity of cell nuclei in this cancer. The fractal curve also describes the way in which the length measurement between two points increases as the scale decreases (Costa et al. 2013; Hernández and Ventura 2022). Several studies have shown the use of fractal methods to describe complex pathological structures including colon, prostate and breast cancer for diagnosis, staging and prognosis (Lee et al. 2014; da Silva et al. 2021). The data showed consistent correlations, providing strong evidence that macroscopic fractal dimension and lacunarity measurements reflect differences in the tumor cells that give rise to them (Mambetsariev et al. 2019). Fractal geometry could give insights into tumor morphology and could become a useful tool for analyzing irregular tumor growth patterns (Mincione et al. 2015). According to Namazi and Kiminezhadmalaie (2015), the damaged DNA exhibits higher degree of fractality compared to normal DNA in lung cancer.

Some studies have shown the ability of the fractal dimension to detect changes in parameters of certain types of lung cancer. According to Lee et al. (2014), the fractal dimension of carcinoma epithelial architecture stained with pancytokeratin can differentiate adenocarcinoma from squamous cell carcinoma of the lung. Mambetsariev et al. (2019) showed an increased fractal dimension in microscopic images and immunohistochemistry (C-MET, P-C-MET, FAK, P-FAK) for small cell lung cancer. However, we did not find previous studies showing the efficiency of fractal dimension in detecting changes in the spatial distribution of cell nuclei in lung SqCC, as shown in the present study.

According to Lee et al. (2014), the use of fractal dimension to differentiate lung cancer subtypes is a nascent technology, but studies with a large number of cases are still needed. In this way, the present study can contribute to consolidating the use of the fractal dimension as a large number of images (400) from a database were used in our methodology. It is worth mentioning that the fractal dimension has also been used recently to detect alterations in other non-cancer lung pathologies. Mancini et al. (2023) provided evidence to support the use of fractal morphometry as a tool for quantifying and determining lung tissue remodeling in idiopathic pulmonary fibrosis. It is also worth highlighting the increase in the fractal dimension of cell nuclei in carcinoma of the oral mucosa (Mincione et al. 2015).

The theoretical characterization of our images as a fractal structure may represent a limitation; but this possible limitation often occurs in the mathematical use of the method in practical situations applied to pathology. According to Lee et al. (2014), it should be noted that actual fractals do not exist in nature because there is a fundamental limitation to the scaling behavior of natural objects. Even real image renderings of mathematical fractals cannot be truly fractal because of the finite resolution of the rendering. Hence, as in previous studies, the concept of fractal dimension refers to the complexity of an object over a very finite range of scales, and thus represents a pragmatic tool to extract clinically meaningful information.

The lacunarity method measures the spatial distribution of gaps in the structure, quantifying the homogeneity of an image or part of it, in order to make it comparable with other images and quantifying the space filled in this structure. Lacunarity describes the heterogeneity of the structure, the higher the value of the lacunarity the greater the heterogeneity (Melo et al. 2022). According to Mambetsariev et al. (2019), lacunarity also analyzes the texture within the image beyond what can be observed with standard measurements, making it ideal to quantify the complexity of stained tissue.

Lacunarity analysis in the present study showed a statistical reduction (41%) in the SqCC; interestingly, even though lacunarity is less studied in lung cancer compared to fractal dimension, it showed a greater change than the fractal dimension (28%) in the SqCC images. The use of lacunarity in lung cancer microscopy is still scarce; however, the present study showed lacunarity as a useful tool applied to lung pathology. Mambetsariev et al. (2019) described lacunarity as a method capable of differentiating benign from malignant tissue and also providing additional information on the biology of the tumor. These authors showed reduced lacunarity in microscopic images and immunohistochemistry (C-MET, P-C-MET, FAK, P-FAK) for small cell lung cancer. Studies analyzing the lacunarity of cell nuclei in microscopic images of lung cancer are scarce, and we found no previous reports using lacunarity to analyze the distribution of cell nuclei in human lung SqCC.

Very few studies have used multifractal to evaluate lung cancer. The principle behind multifractal analysis is that diferent regions of a process may have different fractal properties, providing information on a broad range of heterogenous phenomena (Barbora et al. 2023). These authors used multifractal analysis on acquired FTIR spectra to demonstrate the cancer metastatic potential in B16-f01, B16-f10, SW-480 and SW-620 cancer cell lines. The multifractal describes the structure analyzed at different scales; this method may be more efficient to evaluate biological structures that have different characteristics at different scales and dimensions (Melo et al. 2022). The present study showed a statistical increase of the parameters Dq, α and f(α) for all values of q (-10 to + 10), with a greater increase for more positive q values. Therefore, multifractality was able to detect changes in the spatial distribution of cell nuclei in lung SqCC. It is worth mentioning that the complex shape marker with the greatest change observed in this study was the f(α) for q + 10, which increased by 53% in lung SqCC images. Surprisingly, studies using multifractal methods are even rarer in health sciences, so we found no previous reports using multifractality to analyze the distribution of cell nuclei in human lung SqCC.

Carcinogenesis involves, among other characteristics, impaired cellular proliferation control and the capacity to invade tissue (da Silva et al. 2021). In this sense, the results observed in the present study for traditional measures (number of cells, area and perimeter) corroborate the literature. The changes observed in our results for simpler shape markers: circularity, aspect ratio, roundness and solidity, corroborate the data observed by Wang et al. (2014). Even so, it is worth highlighting that descriptions using these parameters in SqCC images are still few.

Traditional shape markers in microscopy may have limitations depending on the characteristics of the structure. For example, the area is one of the most analyzed parameters in images, however, the area only provides information on the amount of space the structure occupies, but does not inform how this structure occupies the image. In a simple example, the cell can show a circle shape with area 100 in benign tissue and the same area 100 with a shape similar to a star in a pathology. Furthermore, the area of the nucleus (2 dimensions) can vary greatly depending on the shape and the angle of the cross-section because the nucleus in vivo has 3 dimensions. Even so, optical microscopy is still one of the most used methods to evaluate cancer, despite its limitation when analyzing 2-dimensional shape of 3-dimensional structures, resulting in great variability. In this sense, proposing complex markers capable of generating new information about how cell nuclei occupy the image (not just the amount) can bring new possibilities in detecting and understanding the development of cancer.

Given the rising availability and lower pricing of digital cameras and slide scanners, the use of fractal methods could be easily added in the diagnostic work-up of both surgical and small biopsy samples. Moreover, it could be implemented in artificial intelligence-guided pattern recognition, helping to provide a more comprehensive approach for difficult cases (Mancini et al. 2023). The present study analyzed a human database with a large number of images and diverse characteristics. Thus, the complex shape markers applied in the present study showed good robustness and efficiency independent of adverse factors such as sample collection, processing, coloring, intensity, lighting, among other variations present in images from human databases (outside controlled conditions of experiments in laboratory). Studies have been analyzing the LC25000 dataset to automate cancer detection (Al-Jabbar et al. 2023; Dabass et al. 2023; Gabralla et al. 2023; Halder and Dey 2023). However, most use methods involving machine learning, neural networks, deep learning and other AI-related protocols. On the other hand, the present study followed a different path, aiming to use the LC25000 dataset in search of complex shape markers to better describe the spatial distribution of cells in lung cancer. We did not find previous studies in this approach evaluating the database using fractal dimension, lacunarity and multifractality. It is worth mentioning that complex shape markers can be easily integrated into AI protocols.

Methods analyzing complex shape markers for automatic detection of lung SqCC are still rare, and analyzing the fractal dimension, lacunarity and especially multifractality in the spatial distribution of cell nuclei are practically absent. Therefore, the results in the present study may indicate new tools applicable to the development of computational methods and artificial intelligence for SqCC detection. In association with a pathologist to supervise the diagnosis, automated methods using complex shape markers could be useful new tools in the near future to increase the efficiency and speed of analysis, reducing costs by serving as a first or second analyzer in cancer detection. However, further studies are still needed to determine the applicability of these methods in practical routine evaluating different classifications of SqCC and different types of lung cancer.

DECLARATION OF INTEREST

The authors report no declarations of interest.

FUNDING

This work was supported by the Federal University of Pernambuco, Recife, Pernambuco, Brazil.

Author Contribution

Histopathological analysis: Santos AVF, Tenorio FCAM, Junior VASJ; Quantitative morphometry: Santos AVF, Maia, CS, Tenorio FCAM; Fractal dimension, lacunarity and multifractality analyzes: Santos AVF, Tenorio BM, Nogueira RA; Interpretation of data: Tenorio BM, Nogueira RA, Maia, CS; Wrote the manuscript: Junior VASJ, Nogueira RA, Tenorio BM; Coordinated the research: Tenorio BM.

5. ACKNOWLEDGMENTS

The authors thank Borkowski et al. (2019B) for the LC25000 database.

Adams SJ, Stone E, Baldwin DR, Vliegenthart R, Lee P, Fintelmann FJ (2023) Lung cancer screening. Lancet 4;401:390-408. doi: 10.1016/S0140-6736(22)01694-4.
Al-Jabbar M, Alshahrani M, Senan EM, Ahmed IA (2023) Histopathological Analysis for Detecting Lung and Colon Cancer Malignancies Using Hybrid Systems with Fused Features. Bioengineering 10(3):383. doi: 10.3390/bioengineering10030383.
Barbora A, Karri S, Firer MA, Minnes R (2023) Multifractal analysis of cellular ATR-FTIR spectrum as a method for identifying and quantifying cancer cell metastatic levels. Sci Rep 2;13(1):18935. doi: 10.1038/s41598-023-46014-1.
Borkowski AA, Wilson CP, Borkowski SA, Thomas LB, Deland LA, Grewe SJ, Mastorides SM (2019) Comparing Artificial Intelligence Platforms for Histopathologic Cancer Diagnosis. Fed Pract A 36(10):456-463. PMID: 31768096.
Borkowski AA, Bui MM, Thomas LB, Wilson CP, DeLand LA, Mastorides SM (2019) Lung and Colon Cancer Histopathological Image Dataset (LC25000). arXiv 1912.12142v1 [eess.IV].
Casey DT, Lahue KG, Mori V, Herrmann J, Hall JK, Suki B, Janssen-Heininger YMW, Bates JHT (2024) Local fractal dimension of collagen detects increased spatial complexity in fibrosis. Histochem Cell Biol. 161(1):29-42. doi: 10.1007/s00418-023-02248-8.
Costa EVL, Jimenez GC, Barbosa CTF, Nogueira RA (2013) Fractal analysis of extra-embryonic vascularization in Japanese quail embryos exposed to extremely low frequency magnetic Fields. Bioelectromagnetics. 34:114-121. doi: http://dx.doi.org/10.1002/bem.21759
Costa EVL, Nogueira RA (2015) Multifractal dimension and lacunarity of yolk sac vasculature after exposure to magnetic ﬁeld. Microvasc Res 99, 1–7. http://dx.doi.org/10.1016/j.mvr.2015.02.002.
Dabass M, Dabass J, Vashisth S, Vig RA (2023) hybrid U-Net model with attention and advanced convolutional learning modules for simultaneous gland segmentation and cancer grade prediction in colorectal histopathological images. Intell Based Med. 7:100094. https://doi.org/10.1016/j.ibmed.2023.100094.
da Silva LG, da Silva Monteiro WRS, de Aguiar Moreira TM, Rabelo MAE, de Assis EACP, de Souza GT (2021) Fractal dimension analysis as an easy computational approach to improve breast cancer histopathological diagnosis. Appl Microsc 30;51(1):6. doi: 10.1186/s42649-021-00055-w.
Gabralla LA, Hussien AM, AlMohimeed A, Saleh H, Alsekait DM, El-Sappagh S, Ali AA, Refaat Hassan M (2023) Automated Diagnosis for Colon Cancer Diseases Using Stacking Transformer Models and Explainable Artificial Intelligence. Diagnostics 13(18):2939. doi: 10.3390/diagnostics13182939.
Halder A, Dey D (2023) MorphAttnNet: An Attention-based morphology framework for lung cancer subtype classification. Biomed Signal Process Control 86:105149. https://doi.org/10.1016/j.bspc.2023.105149
Hernandez GR, Ventura PM (2022) The neuroglia of the optic nerve. Part II. Fractal morphometry of astrocytes. Eur J Anat 26, 179–187. https://doi.org/10.52083/URGQ2358.
Lee LH, Tambasco M, Otsuka S, Wright A, Klimowicz A, Petrillo S, Morris D, Magliocco A, Bebb DG (2014) Digital differentiation of non-small cell carcinomas of the lung by the fractal dimension of their epithelial architecture. Micron 67:125-131. doi: 10.1016/j.micron.2014.07.005.
Leiter A, Veluswamy RR, Wisnivesky JP (2023) The global burden of lung cancer: current status and future trends. Nat Rev Clin Oncol 20(9):624-639. doi: 10.1038/s41571-023-00798-3.
Lennon FE, Cianci GC, Cipriani NA, Hensing TA, Zhang HJ, Chen CT, Murgu SD, Vokes EE, Vannier MW, Salgia R (2015) Lung cancer-a fractal viewpoint. Nat Rev Clin Oncol 12(11):664-75. doi: 10.1038/nrclinonc.2015.108.
Mambetsariev I, Mirzapoiazova T, Lennon F, Jolly MK, Li H, Nasser MW, Vora L, Kulkarni P, Batra SK, Salgia R (2019) Small Cell Lung Cancer Therapeutic Responses Through Fractal Measurements: From Radiology to Mitochondrial Biology. J Clin Med 16;8(7):1038. doi: 10.3390/jcm8071038.
Mancini M, Bargiacchi L, De Vitis C, D'Ascanio M, De Dominicis C, Ibrahim M, Rendina EA, Ricci A, Di Napoli A, Mancini R, Vecchione A (2023) Histologic Analysis of Idiopathic Pulmonary Fibrosis by Morphometric and Fractal Analysis. Biomedicines. 19;11(5):1483. doi: 10.3390/biomedicines11051483.
Melo IO, Angelo Mendes Tenorio FDC, da Silva Gomes JA, da Silva Junior VA, de Albuquerque Nogueira R, Tenorio BM (2022) Fractal methods applied to the seminiferous lumen images can quantify testicular changes induced by heat stress. Acta Histochem 124(7):151949. doi: 10.1016/j.acthis.2022.151949.
Mincione G, Di Nicola M, Di Marcantonio MC, Muraro R, Piattelli A, Rubini C, Penitente E, Piccirilli M, Aprile G, Perrotti V, Artese L (2015) Nuclear fractal dimension in oral squamous cell carcinoma: a novel method for the evaluation of grading, staging, and survival. J Oral Pathol Med. 44(9):680-4. doi: 10.1111/jop.12280.
Namazi H, Kiminezhadmalaie M (2015) Diagnosis of Lung Cancer by Fractal Analysis of Damaged DNA. Comput Math Methods Med 242695. doi: 10.1155/2015/242695.
Paech DC, Weston AR, Pavlakis N, Gill A, Rajan N, Barraclough H, Fitzgerald B, Van Kooten M (2011) A systematic review of the interobserver variability for histology in the differentiation between squamous and nonsquamous non-small cell lung cancer. J Thorac Oncol 6(1):55-63. doi: 10.1097/JTO.0b013e3181fc0878.
Pala MA, Çimen ME, Akgül A et al. (2022) Fractal dimension-based viability analysis of cancer cell lines in lens-free holographic microscopy via machine learning. Eur Phys J Spec Top 231:1023-1034. https://doi.org/10.1140/epjs/s11734-021-00342-3
Sánchez-Chávez HD, Flores-Cano L, Santiago-Hernández A, García-Soriano AA (2024) Multifractal approach for a biological porous media: Human dentin case. Microsc Res Tech 87(1):10-20. doi: 10.1002/jemt.24378.
Tenorio BM, da Silva Filho EA, Neiva GSM, da Silva VA, Tenorio FDCAM, da Silva TJ, Silva ECSE, Nogueira RA (2017) Can fractal methods applied to video tracking detect the effects of deltamethrin pesticide or mercury on the locomotion behavior of shrimps? Ecotoxicol Environ Saf 15(142):243-249. doi: 10.1016/j.ecoenv.2017.03.051.
Wang H, Xing F, Su H, Stromberg A, Yang L (2014) Novel image markers for non-small cell lung cancer classification and survival prediction. BMC Bioinformatics 19;15(1):310. doi: 10.1186/1471-2105-15-310.
World Health Organization (WHO). Lung cancer. 2023. https://www.who.int/news-room/fact-sheets/detail
Yatabe Y, et al. (2019) Best Practices Recommendations for Diagnostic Immunohistochemistry in Lung Cancer. J Thorac Oncol 14(3):377-407. doi: 10.1016/j.jtho.2018.12.005.

No competing interests reported.

Download PDF

Editorial decision: Revision requested
25 Jul, 2024
Reviews received at journal
25 Jul, 2024
Reviews received at journal
23 Jul, 2024
Reviewers agreed at journal
22 Jul, 2024
Reviews received at journal
18 Jul, 2024
Reviewers agreed at journal
15 Jul, 2024
Reviews received at journal
13 Jul, 2024
Reviewers agreed at journal
08 Jul, 2024
Reviewers agreed at journal
08 Jul, 2024
Reviewers invited by journal
08 Jul, 2024
Editor assigned by journal
06 Jul, 2024
Submission checks completed at journal
06 Jul, 2024
First submitted to journal
05 Jul, 2024

You are reading this latest preprint version

Complex shape markers can detect alterations in the spatial distribution of cell nuclei in human lung squamous cell carcinoma: a useful tool for automatic analysis?

Status:

Version 1

Abstract

Figures

1. INTRODUCTION

2. MATERIALS AND METHODS

2.1 EXPERIMENTAL DESIGN

2.2 FRACTAL DIMENSION

2.3 LACUNARITY

2.4 MULTIFRACTALITY

2.5 QUANTITATIVE MORPHOMETRY

2.6 STATISTICAL ANALYSIS

3. RESULTS

3.1 HISTOPATHOLOGICAL ANALYSIS

3.2 FRACTAL DIMENSION AND LACUNARITY

3.3 MULTIFRACTALITY

3.4 QUANTITATIVE MORPHOMETRY

4. DISCUSSION

Declarations

DECLARATION OF INTEREST

FUNDING

Author Contribution

5. ACKNOWLEDGMENTS

References

Additional Declarations

Status:

Version 1