Fractal Dimension, Circularity, and Solidity of Cell Clusters in Liquid-Based Endometrial Cytology Are Potentially Useful for Endometrial Cancer Detection and Prognosis Prediction

Abstract

Endometrial cancer (EC) in women is increasing globally, necessitating improved diagnostic methods and prognosis prediction. While endometrial histology is the conventional approach, liquid-based endometrial cytology may benefit from novel analytical techniques for cell clusters. A clinical study was conducted at the University of Fukui Hospital from 2012 to 2018, involving 210 patients with endometrial cytology. The liquid-based cytology images were analyzed using cell cluster analysis with Image J software. Logistic regression, ROC analysis, and survival analysis were employed to assess the diagnostic accuracy and prognosis between cell cluster analysis and EC/atypical endometrial hyperplasia (AEH). Circularity and fractal dimension demonstrated significant associations with EC and AEH, regardless of age and cytology results. The ROC analysis revealed improved diagnostic accuracy when combining fractal dimension with cytology, particularly in menopausal age groups. Lower circularity and solidity were independently associated with poor overall survival, while higher fractal dimension values correlated with poorer overall survival in Grades 2 and 3 endometrial cancers. The combination of circularity and fractal dimension with cytology improved diagnostic accuracy for both EC and AEH. Moreover, circularity, solidity, and fractal dimension may serve as prognostic indicators for endometrial cancer, contributing to the development of more refined screening and diagnostic strategies.

Keywords: computer assisted; cytodiagnosis; cytological techniques; endometrial neoplasms; fractals; image processing; prognosis.

Figure 1

Processing of a liquid-based endometrial cytology image. (A,B) A bright image of the normal endometrium or endometrial cancer was obtained using a 4× objective lens. It was then converted to an 8-bit binary image using ImageJ software.

Figure 2

ROC curve analysis for all ages. The numbers on the graph represent following value: Cutoff (Specificity, Sensitivity). ROC curves for the detection of AEH and EC (EM G1, EM G2, EM G3, serous cancer, and clear cell cancer) based on the following parameters: (A) perimeter, (B) fit ellipse major, (C) fit ellipse minor, (D) angle, (E) circularity, (F) Max Feret diameter, (G) Min Feret diameter. (H) solidity, and (I) fractal dimension. Endometrioid cancer (EM); grade (G); atypical endometrial hyperplasia (AEH); receiver operating characteristic (ROC); maximum (Max); minimum (Min).

Figure 3

ROC curve analysis for ages over 55 yr. The numbers on the graph represents the following value: Cutoff (Specificity, Sensitivity). ROC curves for the detection of AEH and EC (EM G1, EM G2, EM G3, serous cancer, and clear cell cancer) based on the following parameters: (A) perimeter, (B) fit ellipse major, (C) fit ellipse minor, (D) angle, (E) circularity, (F) Max Feret diameter, (G) Min Feret diameter, (H) solidity, and (I) fractal dimension. Endometrioid cancer (EM); grade (G); atypical endometrial hyperplasia (AEH); receiver operating characteristic (ROC); maximum (Max); minimum (Min).

Figure 4

Performance analysis of cell clusters with cytology for detecting the EC and AEH. EC includes endometrioid cancer, serous cancer, and clear cell cancer. The EC includes endometrioid cancer, serous cancer, and clear cell cancer. The AUC of the ROC curve for cytology diagnosis was compared to that for cytology diagnosis plus circularity, solidity, and fractal dimension in (A,C,E) for all ages and (B,D,F) for ages > 55 yr, respectively, atypical endometrial hyperplasia (AEH), endometrial cancer (EC), and CY: cytology diagnosis.

Figure 5

Prognostic impact of circularity. Cutoffs were set using X-tile software. EC includes EM, serous cancer, and clear cell cancer. (A–C) Lower circularity was significantly associated with poor OS in EC and EM as well as EM G2 and G3. (A) EC; (B) EM; (C) EM G2 and G3. Endometrial cancer (EC); endometrioid cancer (EM); grade (G).

Figure 6

Prognostic impact of solidity. Cutoffs were set using X-tile software. EC includes EM, serous cancer, and clear cell cancer. (A–C) Lower solidity was significantly associated with poor OS in EC and EM as well as EM G2 and G3. (A) EC; (B) EM; (C) EM G2 and G3. Endometrial cancer (EC); endometrioid cancer (EM); grade (G).

Figure 7

Prognostic impact of fractal dimensions. Cutoffs were set using X-tile software. EC includes EM, serous cancer, and clear cell cancer. (A,B) The results showed no association between OS and fractal dimension for EC and EM. (C) Higher fractal dimension was significantly associated with poor OS in EM G2 and G3. (A) EC; (B) EM; (C) EM G2 and G3. Endometrial cancer (EC); endometrioid cancer (EM); grade (G).