In this blog we continue analyzing the importance of color accuracy in dermoscopy, considering multispectral dermoscopy, color constancy and consistency, greyscale and color resolution, and the use of smartphones in dermatology.
Multispectral dermoscopy
Enhanced performances in the analysis of skin lesions could be obtained by using multispectral dermoscopy, based on the illumination of the skin with narrowband light sources with different wavelengths. It could be a way to enhance the visualization of vasculature and pigment. Each of the wavelengths is differently absorbed by skin chromophores, such as pigment or (de)oxygenated blood. A study[1] investigated the additional information provided by such “skin parameter maps” in some cases of basal cell carcinoma and Bowen's disease; the conclusions were that skin parameter maps based on multispectral images can give better insight in the inner structures of lesions, especially in lesions with characteristic blood vessels.
Color constancy and consistency
In addition to fidelity of color reproductions another important feature in dermatology is represented by color constancy. Differences in illumination conditions and setups, differences in acquisition devices alter the color of images and it could affect both the visual assessment by the dermatologists and the outcomes of computer-aided diagnosis systems.
It will be difficult to ensure the robustness of CAD systems if they operate with multisource images acquired under different setups. Thus, it is important to normalize the colors of dermoscopy images.
A recently published paper[2] investigated different constancy algorithms (Gray World, max-RGB, Shades of Gray, and General Gray World) and confirmed the improvement of the classification of multisource images when a crtain level of color constancy is ensured.
Image consistency is needed in sequential dermoscopy imaging (SDI) and this is a critical issue since unfortunately differences in technologies and different imaging devices can produce different results.
In this regard the lack of standards in dermoscopy represents a key barrier.
Greyscale and color resolution
Another topic of research in dermoscopy is the use of High Dynamic Range for the enhancement in the visualization of some dermoscopic structures[3]. The pigmented structures at the periphery are more conspicuous in HDR mode and can easily be identified as spoke wheel like structures and leaf-like areas. The blood vessels are also rendered more conspicuous and are identified as arborizing telangiectasia. Furthermore, the crystalline structures in the center of the lesion become more noticeable. As a result, the diagnosis of basal cell carcinoma is more obvious in the dermoscopic HDR image.
The use of smartphones in dermatology
The topic of image quality in dermoscopy is becoming even more critical in consideration of the growing use of smartphones by patients and by dermatologists.
A recently published paper[4] compared the skin images obtained with a smartphone with those of a reference dermatoscope. Results indicate the superiority of these last ones with regard to image clarity as well as observed color fidelity. Smartphone sensors showed areas with saturated pixels, as opposed to the reference device. The reason is that often in consumer-based cameras and devices the image processing software is designed to generate more saturated colors, pleasant to the broadest population of users.
An additional concern is represented by the image sharpening process adopted in smartphones; typically a high level of sharpening is used and this amplifies image noise and image sampling artifacts[5] .
Wrong color and too intensive sharpening can produce artifacts or border contour changes, which can result in less effective melanoma screening, longer diagnostic times and potentially wrong decisions[6].
New smartphones are launched in a rapid pace, often using different chipsets and optics; furthermore, software updates can result in (unnoticed but) different image quality that may influence the output of AI algorithms. A recently published paper investigated how AI copes with changes in image acquisition[7]
[1] Laudine Janssen, Sofie Mylle, Sofie Van Kelst, Julie De Smedt, Bart Diricx, Tom Kimpe, Marc Boone, Evelien Verhaeghe, Lieve Brochez, Marjan Garmyn - Enhanced visualization of blood and pigment in multispectral skin dermoscopy; Skin Research & Technology – Volume 26, Issue 5 September 2020 - Pages 708-712
[2] C. Barata, M. E. Celebi, and J. S. Marques - Improving Dermoscopy Image Classification Using Color Constancy - IEEE JOURNAL OF BIOMEDICAL AND HEALTH INFORMATICS, VOL. 19, NO. 3, MAY 2015
[3] Braun RP, Marghoob A. – A high dynamic range dermoscopy imaging and diagnosis of hypopigmented skin cancers – JAMA Dermatol 2015; 151:456-457
[4] Varun Vasudev, Lode De Paepe, Andrew D. A. Maidment, Tom Kimpe, Ljiljana Platisa, Wilfried Philips, and Predrag R. Bakic "Effects of smartphone sensor characteristics on dermatoscopic images: a simulation study", Proc. SPIE 11595, Medical Imaging 2021: Physics of Medical Imaging, 115952I (15 February 2021); https://doi.org/10.1117/12.2582043
[5] B. Dugonik et al. – Image quality assessment of digital image capturing devices for melanoma detection – Appl. Sci. 2020, 10, 2876, doi:10.3390/app 10082876; www.mdpi.com/journal/applsci
[6] Rosado L. et al.- From dermoscopy to mobile teledermatology – Dermoscopy Imaging Analysis; CRC Press: Boca Raton, Fl, USA, 2015.
[7] J. de Vylder, T. Kimpe – The impact of consistent image capturing for AI-based systems – JAAD Journal of the Americal Academy of Dermatology - DOI:https://doi.org/10.1016/j.jaad.2020.06.243