Multiclass ML Classification for Brain Metastases Distribution

Brain metastases (BM) present a significant clinical challenge due to their association with poor prognosis and increased mortality rates. Understanding BMs is crucial for enhancing early detection and monitoring in brain cancer patients. Despite the importance of studying the anatomical distributions of BM from various primary cancers, comprehensive systematic comparisons in this area are currently lacking.

Saeedeh Mahmoodifar and the team conducted a study that aimed to conduct systematic comparisons of the anatomical distributions of BM from various primary cancers.

In the analysis, the team tested the hypothesis that anatomical distributions of BM differ based on primary cancer type. They examined the spatial coordinates of BMs from five distinct primary cancer types along principal component (PC) axes.

The dataset, comprising 3949 intracranial metastases labeled by primary cancer types and featuring 6 characteristics, was utilized. PC coordinates were leveraged to accentuate differences between cancer types. Various Machine Learning (ML) algorithms, including Random Forest (RF), Support Vector Machine (SVM), and TabNet Deep Learning (DL), were employed to establish relationships between primary cancer diagnosis, spatial coordinates of BMs, age, and target volume.

The findings indicated that PC1 aligns predominantly with the Y axis, followed by the Z axis, and exhibited minimal correlation with the X axis. Upon examining PC1 versus PC2 plots, notable differences in anatomical spreading patterns were observed between Breast and Lung cancer, also in Breast and Renal cancer. Conversely, Renal and Lung cancer, along with Lung and Melanoma, displayed similar patterns.

Results from ML and DL approaches showcased high accuracy in distinguishing BM distribution for various primary cancers, with the SVM algorithm achieving 97% accuracy using a polynomial kernel and TabNet achieving 96%. The RF algorithm identified PC1 as the most significant discriminating feature.

The findings substantiate the effectiveness of multiclass ML classification in accurately characterizing brain metastases distribution.

Partial funding was provided by the USC Norris Comprehensive Cancer Center.

Source: https://link.springer.com/article/10.1007/s11060-024-04630-5

Mahmoodifar, S., Pangal, D.J., Neman, J. et al. “Comparative analysis of the spatial distribution of brain metastases across several primary cancers using machine learning and deep learning models.” J Neurooncol (2024). https://doi.org/10.1007/s11060-024-04630-5.