Machine learning-based histological classification that predicts recurrence of peripheral lung squamous cell...

Area of Science:

• Oncology
• Computational Pathology
• Biostatistics

Background:

• Lung squamous cell carcinoma (SqCC) comprises cancer and stromal cells.
• Tumor microenvironment composition may influence patient outcomes.

Purpose of the Study:

• To investigate if the cancer cell to stromal component ratio predicts prognosis in SqCC patients.
• To utilize machine learning for objective component analysis and prognostic evaluation.

Main Methods:

• Machine learning was employed to quantify cancer cell, necrotic, and stromal components in 135 SqCC cases.
• Cases were classified into predominant cancer cell, necrosis, or stroma subtypes.
• Prognostic significance of subtypes was assessed, focusing on recurrence-free survival (RFS).

Main Results:

• The predominant stroma subtype (70 cases) showed significantly shorter 5-year RFS (42.3%) compared to the predominant cancer cell subtype (59 cases, 84.3%).
• This association remained significant in pathological stage I patients (5-year RFS: 64.3% vs. 88.4%).
• Multivariate analysis confirmed the predominant stroma subtype as an independent prognostic factor for RFS in stage I SqCC.

Conclusions:

• The predominant stroma subtype is an independent negative prognostic factor for RFS in SqCC.
• The ratio of stromal component correlates with the malignant potential of SqCC.
• Machine learning provides objective quantification for prognostic assessment in lung cancer.