The effect of decreasing digital image resolution on teledermatology diagnosis.

Abstract

Objective: To determine the effect of degraded digital image resolution (as viewed on a monitor) on the accuracy and confidence of dermatologic interpretation.

Materials and methods: Eight dermatologists interpreted 180 clinical cases divided into three Logical Competitor Sets (LCS) (pigmented lesions, non-pigmented lesions, and inflammatory dermatoses). Each case was digitized at three different resolutions. The images were randomized and divided into (9) 60-image sessions. The physicians were completely blinded concerning the image resolution. After 60 seconds per image, the viewer recorded a diagnosis and level of confidence. The resultant ROC curves compared the effect of LCS, level of clinical difficulty, and resolution of the digital image. One-way analysis of variance (ANOVA) compared the curves.

Results: The areas beneath the ROC curves did not demonstrate any consistently significant difference between the digital image resolutions for all LCS and levels of difficulty. The only significant effect observed was amongst pigmented lesions (LCS-A) where the ROC curve area was significantly smaller in the easy images at high resolution compared to low and medium resolutions. For all other ROC curve comparisons within LCS-A, at all other levels of difficulty, as well as within the other LCS at all levels of difficulty, none of the differences was significant.

Conclusion: A 720 x 500 pixel image can be considered equivalent to a 1490 x 1000 pixel image for most store-and-forward teledermatology consultations.