Accurate burn severity classification via texture analysis and supervised learning
Abstract
This paper presents a methodology for adapting machine learning (ML) algorithms to determine an optimal approach to burn severity classification and improving classification accuracy. The dataset for multiclass classification includes 1,500 images evenly distributed into three burn severity classes, and the dataset for binary classification includes 1,000 images evenly distributed into two classes. All images were processed with the MaZda 4.6 software package, which computed 297 texture features for each region of interest (ROI) based on different statistical descriptors. In a preprocessing step, the images were converted to 8-bit grayscale format and normalized for luminance. ROIs were manually selected with expert consultation to capture representative texture regions of burn tissue. Six feature selection methods were applied to identify the most informative descriptors. Subsequently, more than ten ML models were trained based on these features. The combination of a multilayer perceptron (MLP) classifier and recursive feature elimination (RFE) yielded the highest multiclass classification accuracy of 79% with 11 selected features. For binary classification, the MLP classifier combined with RELIEF achieved an accuracy of 96% with eight selected features. These results demonstrate the potential of feature selection combined with MLP for burn severity classification.
Keywords
Burn injuries; Burn severity classification; Feature selection; Machine learning; Texture analysis
Full Text:
PDFDOI: https://doi.org/10.11591/eei.v15i6.11751
Refbacks
- There are currently no refbacks.

This work is licensed under a Creative Commons Attribution-ShareAlike 4.0 International License.
Bulletin of Electrical Engineering and Informatics (BEEI)
ISSN: 2089-3191
,
e-ISSN: 2302-9285
This journal is published by the
Institute of Advanced Engineering and Science (IAES)
.