Comprehensive Summary
NHOC (normalized distance from the hot-spot to the tumor centroid) and NHOP (normalized distance from the hot-spot to the tumor perimeter) are metrics of tumor aggressiveness used in PET and CT scanning. This study utilized NHOC and NHOP to determine the likelihood of lymph node metastasis (LNM) in non-small cell lung cancers and then used this data to develop a machine learning model, which was also trained on previous clinical cases as well as PET radiomics for LNM and occult nodal metastasis (ONM) predictions. The multi-level perceptron classifier (MPC), which was the system developed, significantly outperformed previous clinical and NHOC models with statistical significance. It was determined that the MPC model demonstrated superior performance in comparison to the other models for LNM prediction during training (AUC 0.852), internal test (AUC 0.822), and external test (AUC 0.885) sets. When analyzing the MPC model’s capacity for ONM prediction, the combined model achieved the AUC (0.85) on the full datasets.
Outcomes and Implications
This research demonstrates a clear future in AI-oriented analysis of medical imaging, particularly within non-small cell lung cancers. The usage of these systems to determine likelihood of metastasis is invaluable information to both the patient and the physician. However, there is a rather major caveat to this research: while it is claimed that this model demonstrates statistical significance in accuracy, no data is shown in correlation to the other systems to support this assertion.