An outlier detection algorithm based on segmentation and pruning of competitive network for glioma identification...

Area of Science:

• Biomedical Optics
• Spectroscopy
• Machine Learning

Background:

• Raman spectroscopy offers non-invasive, high-information-density analysis for brain glioma diagnosis.
• Identifying glioma patterns is challenging due to tissue heterogeneity and potential outliers from experimental variability.
• Existing outlier detection methods can reduce sample utilization and compromise model robustness.

Purpose of the Study:

• To develop a novel, robust outlier detection algorithm for Raman spectra of brain gliomas.
• To improve the accuracy and reliability of glioma diagnosis using Raman spectroscopy.
• To address challenges posed by high within-group variance and instrument inconsistencies.

Main Methods:

• Proposed the SPCN (Segmentation and Pruning Competitive Network) outlier detection algorithm.
• SPCN utilizes competitive network segmentation, α-β region segmentation, and a two-stage pruning method.
• The algorithm is label-free and does not require outlier distance thresholds or density estimation.

Main Results:

• SPCN demonstrated superior precision and robustness compared to six traditional outlier detection algorithms.
• Analysis of Raman spectra from 113 brain glioma patients confirmed SPCN's effectiveness.
• Outlier removal using SPCN enhanced pattern recognition accuracy in glioma diagnosis.

Conclusions:

• The SPCN algorithm significantly enhances the accuracy and reliability of brain glioma diagnosis via Raman spectroscopy.
• SPCN offers a robust solution for handling outlier spectra in complex biological samples.
• The method is adaptable for outlier detection in other spectral data, including near-infrared and mid-infrared.