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

Abstract

Background: Cancer tissue is composed of both a cancer cell component and a stromal component. The aim of this study was to investigate if the component ratio predicts a prognosis for lung squamous cell carcinoma (SqCC) patients by using a machine learning method.

Methods: A total of 135 peripheral SqCC cases (tumor size: 3-5 cm) were enrolled in this study. The areas of the cancer cell component, the necrotic component, and the stromal component were accurately measured via a machine learning method. Each case was divided into the following three subtypes: 1) predominant cancer cell, 2) predominant necrosis, and 3) predominant stroma. The study examined if a particular subtype had prognostic significance.

Results: The number of cases per subtype of predominant cancer cell, predominant necrosis, and predominant stroma was 59, 6, and 70, respectively. Patients with the predominant stroma subtype had a significantly shorter recurrence free survival (RFS) than did those with the predominant cancer cell subtype (5-yr RFS: 42.3 % vs. 84.3 %,p < 0.01). Also, in pathological stage I patients, the 5-year RFS rate for the predominant stroma subtype was significantly shorter (5-yr RFS: 64.3 % vs. 88.4 %, p < 0.01). In the multivariate analysis of p-stage I patients, the predominant stroma subtype was confirmed to be an independent prognostic factor for RFS (p < 0.01).

Conclusion: Using machine learning, the study confirmed that the predominant stroma subtype was an independent factor for RFS, suggesting that the ratio of the stromal component correlates with the malignant potential of SqCC.

Keywords: Lung squamous cell carcinoma; Machine learning; Recurrence; Stroma.