Dual-Region Computed Tomography Radiomics-Based Machine Learning Predicts Subcarinal Lymph Node Metastasis in Patients with Non-small Cell Lung Cancer

Hao-Ji Yan ^1,2, Jia-Sheng Zhao ³, Hou-Dong Zuo ⁴

⁰Department of Thoracic Surgery, West China Hospital, Sichuan University, Chengdu, China.

Annals of Surgical Oncology +

|

March 23, 2024

View abstract on PubMed

Summary

Predicting subcarinal lymph node metastasis in non-small cell lung cancer (NSCLC) is challenging. A dual-region CT radiomics model using machine learning shows superior accuracy for predicting SLNM in NSCLC patients.

Area Of Science

Medical Imaging
Oncology
Radiology

Background

Accurate prediction of subcarinal lymph node metastasis (SLNM) in non-small cell lung cancer (NSCLC) remains a clinical challenge.
Computed tomography (CT) radiomics offers a noninvasive approach to assess tumor and lymph node characteristics.

Purpose Of The Study

To develop and validate a dual-region CT radiomics model incorporating tumor and subcarinal lymph node (SLN) features.
To predict SLNM in NSCLC patients using machine learning (ML) algorithms.

Main Methods

Retrospective study of 202 NSCLC patients who underwent resection and SLN dissection.
Extraction of radiomic features from preoperative CT scans for both tumor and SLN regions.
Development and internal validation (fivefold cross-validation) of 90 ML models based on single-region and dual-region features, assessed by area under the curve (AUC).

Main Results

Dual-region radiomics models demonstrated superior performance for SLNM prediction (median AUC 0.794) compared to single-region models (tumor AUC 0.746, SLN AUC 0.700).
The highest AUC of 0.880 was achieved by a naive Bayes ML model utilizing dual-region features.
The optimal ML model outperformed the optimal logistic regression model (AUC 0.727).

Conclusions

CT radiomics holds significant potential for accurately predicting SLNM in NSCLC.
Machine learning models incorporating dual-region radiomic features provide enhanced predictive performance over single-region or logistic regression models for SLNM in NSCLC.

Keywords:

Computed tomography Machine learning Non-small cell lung cancer Radiomics Subcarinal lymph node metastasis