Tumor-Infiltrating Lymphocytes in Endometrial Cancer

The study was conducted using properly stored archival formalin-fixed paraffin-embedded tissue blocks. The inclusion criteria required a confirmed diagnosis of EC, adequate quality of archival material, absence of prior neoadjuvant treatment, and complete medical documentation. Tumor staging followed the FIGO classification system based on surgical protocols and pathomorphological examination results. For analytical purposes, patients results were stratified into two groups based on tumor grade: a low-grade group (grade 1 and 2) and a high-grade group (grade 3). Cancer cells and lymphocytes were identified using Hover-Net, a state-of-the-art nucleic segmentation and classification algorithm. Detected cells were categorized into six categories: unlabeled, neoplastic (cancer), inflammatory (TILs, i.e., lymphocytes and plasma cells), connective, necrosis, and non-neoplastic. To estimate cancer areas from cancer cell segmentation masks, a novel block-processing algorithm optimized for large image analysis, was developed. For each tissue sample, the TIL percentage as the area occupied by lymphocytes divided by the cancer area, expressed as a percentage, was calculated. TIL distribution maps were constructed using tissue segmentation masks, cancer region masks, and TIL segmentation masks. Spatial TIL metrics were subsequently calculated based on GLCM analysis and FD. After grey level co-occurrence matrix (GLCM) calculation, different weights were applied to each matrix element to derive two measures: M1 and M2, representing areas with low and high intensities, respectively. Lower M1 and higher M2 values characterized more structured images with distinct TIL patterns. FD provided a statistical index of pattern complexity in geometric structures. A curve with an FD close to 1 resembles an ordinary line (simple structure), while curves with higher FD values exhibit convoluted spatial arrangements resembling spaces. Higher FD values thus indicate more structured and complex TIL distribution patterns.

Data were analyzed using Dell Statistica software v13.3 (TIBCO Software Inc., Palo Alto, California, United States) and MedCalc Statistical Software v19.2.6 (MedCalc Software, Ostend, Belgium).

• Tumor-infliltrating lymphocyte (TIL) percentage (Diagnostic test): TIL percentage was calculated as the area occupied by lymphocytes divided by the cancer area, expressed as a percentage \[%\]
• Grey level co-occurrence matrix (GLCM) (Diagnostic test): The GLCM is a second-order statistical method for texture feature extraction. Structured images typically contain numerous pixel pairs with co-occurring low- and high-intensity values. After GLCM calculation, different weights were applied to each matrix element to derive two measures: M1 and M2, representing areas with low and high intensities, respectively. Lower M1 and higher M2 values characterized more structured images with distinct TIL patterns.
• Fractal dimension (FD) (Diagnostic test): Quantification of the complexity of TIL

Primary outcomes

• TIL percentage [%] [up to 6 months]: TIL percentage is the area occupied by lymphocytes divided by the cancer area, expressed as a percentage \[%\]
• GLCM M1 (×10⁶) [up to 6 months]: GLCM is a second-order statistical method for texture feature extraction. Structured images typically contain numerous pixel pairs with co-occurring low- and high-intensity values. After GLCM calculation, different weights were applied to each matrix element to derive two measures: M1 and M2, representing areas with low and high intensities, respectively. Lower M1 and higher M2 values characterize more structured images with distinct TIL patterns. GLCM M1 values are scaled and expressed in millions (×10⁶), with lower values indicating more structured TIL patterns.
• FD [up to 6 months]: FD provides a statistical index of pattern complexity in geometric structures. Higher FD values thus indicate more structured and complex TIL distribution patterns
• GLCM M2 (×10³) [up to 6 months]: GLCM is a second-order statistical method for texture feature extraction. Structured images typically contain numerous pixel pairs with co-occurring low- and high-intensity values. After GLCM calculation, different weights were applied to each matrix element to derive two measures: M1 and M2, representing areas with low and high intensities, respectively. Lower M1 and higher M2 values characterize more structured images with distinct TIL patterns. GLCM M2 values are scaled and expressed in thousands (×10³), with higher values indicating increased TIL clustering.

Sex: Female, ages: 18 Years to any

Inclusion Criteria:

* confirmed diagnosis of EC
* adequate quality of archival material
* absence of prior neoadjuvant treatment
* complete medical documentation

Exclusion Criteria:

* none

• Milosz Pietrus, PhD, Study chair, Jagiellonian University