Poster Presentation ASPCR-ASDR Conference 2013

Automated location of actinic keratosis from clinical photographs (#95)

Sam Hames , Sudipta Sinnya 1 , Jean-Marie Tan 1 , Peter Soyer 1 , Tarl Prow 1
  1. Dermatology Research Center, School of Medicine, University of Queensland, Brisbane, Queensland, Australia

Actinic keratoses are important precursor lesions associated with the development of squamous cell carcinoma. We propose and test a method of automatically locating actinic keratoses in high quality clinical photographs. The method focuses on detecting the erythema as peaks in a colour space that combines the blue and red chrominance of the input image. Test images were acquired from two groups of six volunteers - one group with severe photodamage and the other group with low photodamage. Images were acquired of each dorsal hand and forearm as well as front, left and right views of the face. Performance was assessed by comparing the automated method with an expert dermatologists assessment of the same digital image files. It was found that there was good colocalisation between the automated method with the automated method matching 65% of the dermatologist identified lesions, though there were significant numbers of false positives, and lesions characterised by scale and horn formation were not identified by the automated method. Comparison of the automatic method on low and high photodamage groups showed a significantly (p<0.05) lower number of lesions in the low photodamage group, indicating that while this method does not ideally match the expert in all cases, automated detection is consistent with the level of photodamage.