Machine Learning-Driven Radiomics for an Early-Stage Predictive Model of Nodal Upstaging in Thoracic Oncology
TL;DR
It is believed that the use of a homogeneous population, however small, can ensure the data uniformity necessary to identify a predictive model that, once refined, can be integrated into clinical practice to personalize the diagnostic pathway and, potentially, the surgical approach.
Abstract
Simple Summary This radiomic study evaluated whether radiomic features from preoperative 2-[18F]FDG PET/CT scans can predict occult lymph node metastases in early-stage non-small cell lung cancer (NSCLC). We enrolled patients with cT1N0 NSCLC with or without occult unexpected node metastasis after surgical resection. Radiomic features of the first and second level were evaluated and machine learning protocol was used to elaborate a predictive model. Both PET/CT and clinical protocols were identical for all patients and were performed at the same institution, providing robust, uniform data with consistent reconstruction parameters, thereby minimizing the data harmonization issue. Despite the inherent limitations of radiomics studies, we believe that the use of a homogeneous population, however small, can ensure the data uniformity necessary to identify a predictive model that, once refined, can be integrated into clinical practice to personalize the diagnostic pathway and, potentially, the surgical approach as well.