Abstract #0021

Investigating machine learning approaches for quality control of brain tumor spectra

Sreenath P Kyathanahally¹, Victor Mocioiu², Nuno Miguel Pedrosa de Barros³, Johannes Slotboom³, Alan J Wright⁴, Margarida Julià-Sapé ², Carles Arús², and Roland Kreis¹

¹Depts. Radiology and Clinical Research, University of Bern, Bern, Switzerland, ²Centro de Investigación Biomédica en Red en Bioingeniería, Biomateriales y Nanomedicina (CIBER-BBN), Universitat Autònoma de Barcelona, Barcelona, Spain, ³DRNN, Institute of Diagnostic and Interventional Neuroradiology/SCAN, University Hospital Bern, Bern, Switzerland, ⁴CRUK Cambridge Institute, University of Cambridge, Cambridge, United Kingdom

Despite many potential applications of MR spectroscopy in the clinic, its usage is limited – and the need for human experts to identify bad quality spectra may contribute to this. Previous studies have shown that machine learning methods can be developed to accept or reject a spectrum automatically. In this study, we extend this to different machine learning methods on 1916 spectra from the eTUMOUR and INTERPRET databases. The RUSBoost classifier, which handles unbalanced data, improved specificity and accuracy compared to other classifiers, in particular in combination with an extended feature set and multi-class labels.

This abstract and the presentation materials are available to members only; a login is required.

Join Here