Malik A S, Humayun J, Kamel N, Yap F B-B
Centre for Intelligent Signal & Imaging Research, Universiti Teknologi Petronas, Tronoh, Malaysia.
Skin Res Technol. 2014 Aug;20(3):322-31. doi: 10.1111/srt.12122. Epub 2013 Dec 12.
More than 99% acne patients suffer from acne vulgaris. While diagnosing the severity of acne vulgaris lesions, dermatologists have observed inter-rater and intra-rater variability in diagnosis results. This is because during assessment, identifying lesion types and their counting is a tedious job for dermatologists. To make the assessment job objective and easier for dermatologists, an automated system based on image processing methods is proposed in this study.
There are two main objectives: (i) to develop an algorithm for the enhancement of various acne vulgaris lesions; and (ii) to develop a method for the segmentation of enhanced acne vulgaris lesions.
For the first objective, an algorithm is developed based on the theory of high dynamic range (HDR) images. The proposed algorithm uses local rank transform to generate the HDR images from a single acne image followed by the log transformation. Then, segmentation is performed by clustering the pixels based on Mahalanobis distance of each pixel from spectral models of acne vulgaris lesions.
Two metrics are used to evaluate the enhancement of acne vulgaris lesions, i.e., contrast improvement factor (CIF) and image contrast normalization (ICN). The proposed algorithm is compared with two other methods. The proposed enhancement algorithm shows better result than both the other methods based on CIF and ICN. In addition, sensitivity and specificity are calculated for the segmentation results. The proposed segmentation method shows higher sensitivity and specificity than other methods.
This article specifically discusses the contrast enhancement and segmentation for automated diagnosis system of acne vulgaris lesions. The results are promising that can be used for further classification of acne vulgaris lesions for final grading of the lesions.
超过99%的痤疮患者患有寻常痤疮。在诊断寻常痤疮皮损的严重程度时，皮肤科医生发现诊断结果存在评分者间和评分者内的差异。这是因为在评估过程中，识别皮损类型及其计数对皮肤科医生来说是一项繁琐的工作。为了使评估工作对皮肤科医生来说更加客观和容易，本研究提出了一种基于图像处理方法的自动化系统。
有两个主要目标：（i）开发一种用于增强各种寻常痤疮皮损的算法；（ii）开发一种用于分割增强后的寻常痤疮皮损的方法。
对于第一个目标，基于高动态范围（HDR）图像理论开发了一种算法。所提出的算法使用局部秩变换从单个痤疮图像生成HDR图像，然后进行对数变换。然后，通过基于每个像素与寻常痤疮皮损光谱模型的马氏距离对像素进行聚类来进行分割。
使用两个指标来评估寻常痤疮皮损的增强效果，即对比度改善因子（CIF）和图像对比度归一化（ICN）。将所提出的算法与其他两种方法进行比较。基于CIF和ICN，所提出的增强算法比其他两种方法都显示出更好的结果。此外，还计算了分割结果的敏感性和特异性。所提出的分割方法比其他方法显示出更高的敏感性和特异性。
本文具体讨论了寻常痤疮皮损自动诊断系统的对比度增强和分割。结果很有前景，可用于寻常痤疮皮损的进一步分类以进行皮损的最终分级。
