作者
Kento Ikuta,Kohei Fukuoka,Yoshiko Suyama,Shunjiro Yagi
摘要
Digital images of the skin play an important role in the clinical evaluation of skin diseases. Improvements in image recognition technology have enabled evaluation of digital images using deep learning. Such technology requires digital images that can be evaluated under appropriate conditions,1 but it is difficult to capture digital images that meet these conditions. It is preferred to use image correction color charts to adjust color, size, brightness, and color temperature when taking case photographs. CasMatch (Bear Medic, Tokyo, Japan) is an image correction color chart commonly used in Japan. It is supplied as a 10 × 10 mm adhesive sticker with nine colors, marked with a 1 mm scale. The color tones can be adjusted to more closely resemble the subject using image processing software, and the 1 mm scale can be used for size measurements. Color correction and size measurements play an important role in image recognition. CasMatch is generally affixed to healthy skin around an area of skin disease (Fig. 1A). It is not suitable for use on uneven surfaces, such as the toes, because misalignment between the CasMatch and the lesion may affect size measurement by image recognition. In addition, when used with elderly patients or patients with skin diseases, the adhesive nature of this product could cause medical adhesive-related skin injury (MARSI).2 The prevalence of MARSI in acute care hospitals has been reported as 3.4%–25.0%, and is a problem in patient care.3 Until now, insufficient attention has been paid to the effective use of image correction color charts. We propose an alternative method in which the CasMatch sticker is attached to a ruler and held in position near the area of interest while the photograph is taken (Fig. 1B).Fig. 1.: Photographs showing the use of CasMatch. A, A CasMatch adhesive sticker is applied to the patient's skin. B, If the sticker should not be applied to the skin due to skin disease, it can be attached to a ruler and positioned nearby.This method has two advantages. First, it can be positioned correctly even in areas where proper application is difficult (such as the toes), thus enabling image recognition to be performed. Second, there is no risk of MARSI. The difference in quality of photographs between the proposed and conventional methods has not yet been examined. However, in our experience, it was possible to take better quality and more suitable case photographs for image recognition than with conventional method, especially in areas with uneven surfaces. It is expected that research in the field of image recognition will continue to grow. The accuracy of deep learning improves with larger sample sizes.4 Awareness of the use of image correction color charts that are appropriate for image recognition will be effective in increasing the number of appropriate case photos. We consider that the present findings will lead to a more feasible environment for future image recognition research. Photographic standards applicable to image recognition will be expected. To summarize, the use of image correction color charts placed on a ruler allows for correct positioning even in areas where proper application is difficult and prevents skin damage. We believe that this report will assist in future research in image recognition technology.