TY - JOUR

T1 - Automated Region Extraction from Thermal Images for Peripheral Vascular Disease Monitoring

AU - Gauci, Jean

AU - Sturgeon, Cassandra

PY - 2018/12/13

Y1 - 2018/12/13

N2 - This work develops a method for automatically extracting temperature data from prespecified anatomical regions of interest from thermal images of human hands, feet, and shins for the monitoring of peripheral arterial disease in diabetic patients. Binarisation, morphological operations, and geometric transformations are applied in cascade to automatically extract the required data from 44 predefined regions of interest. The implemented algorithms for region extraction were tested on data from 395 participants. A correct extraction in around 90% of the images was achieved. The process of automatically extracting 44 regions of interest was performed in a total computation time of approximately 1 minute, a substantial improvement over 10 minutes it took for a corresponding manual extraction of the regions by a trained individual. Interrater reliability tests showed that the automatically extracted ROIs are similar to those extracted by humans with minimal temperature difference. This set of algorithms provides a sufficiently accurate and reliable method for temperature extraction from thermal images at par with human raters with a tenfold reduction in time requirement. The automated process may replace the manual human extraction, leading to a faster process, making it feasible to carry out large-scale studies and to increase the regions of interest with minimal cost. The code for the developed algorithms, to extract the 44 ROIs from thermal images of hands, feet, and shins, has been made available online in the form of MATLAB functions and can be accessed from http://www.um.edu.mt/cbc/tipmid.

AB - This work develops a method for automatically extracting temperature data from prespecified anatomical regions of interest from thermal images of human hands, feet, and shins for the monitoring of peripheral arterial disease in diabetic patients. Binarisation, morphological operations, and geometric transformations are applied in cascade to automatically extract the required data from 44 predefined regions of interest. The implemented algorithms for region extraction were tested on data from 395 participants. A correct extraction in around 90% of the images was achieved. The process of automatically extracting 44 regions of interest was performed in a total computation time of approximately 1 minute, a substantial improvement over 10 minutes it took for a corresponding manual extraction of the regions by a trained individual. Interrater reliability tests showed that the automatically extracted ROIs are similar to those extracted by humans with minimal temperature difference. This set of algorithms provides a sufficiently accurate and reliable method for temperature extraction from thermal images at par with human raters with a tenfold reduction in time requirement. The automated process may replace the manual human extraction, leading to a faster process, making it feasible to carry out large-scale studies and to increase the regions of interest with minimal cost. The code for the developed algorithms, to extract the 44 ROIs from thermal images of hands, feet, and shins, has been made available online in the form of MATLAB functions and can be accessed from http://www.um.edu.mt/cbc/tipmid.

U2 - 10.1155/2018/5092064

DO - 10.1155/2018/5092064

M3 - Journal article

VL - 2018

JO - Journal of Healthcare Engineering

JF - Journal of Healthcare Engineering

SN - 2040-2295

M1 - 5092064

ER -