Digital image analysis for diagnosis of skin tumors
Digital image analysis for diagnosis of skin tumors
复制标题
DOI：
10.1016/j.sder.2007.12.005
复制
发表时间：
2008-03-01
影响因子：
--
通讯作者：
Argenziano, Giuseppe
中科院分区：
文献类型：
--
作者：
Blum, Andreas;Zalaudek, Iris;Argenziano, Giuseppe
在1987至2007年间，不同的组织开发了数字图像分析系统，用于诊断良性和恶性皮肤肿瘤。由于纳入的皮肤肿瘤在技术设备、数量、性质和良恶性比、不同的变量和统计方法上存在显著差异，这些不同的系统及其结果很难进行任何比较。为了将来使用和比较不同数字图像分析系统的诊断性能，需要一些基本的原则条件：所有使用的系统都应该有一个标准化的记录系统和校准。首先，为了开发诊断算法，黑素细胞和非黑素细胞病变应该包括在内。对结果的批判性分析应该回答这样一个问题：将来应该只分析黑色素细胞病变，还是应该分析所有有色素性和非色素性病变。如果只有皮肤科医生或所有内科医生都使用这样的系统，这也会导致答案。所有人工制品(如毛发、气泡)都应该去掉。变量的数量应该根据纳入的黑色素瘤的数量来选择。应该包括大量的良性皮肤病变。在所有病变中，只有10%或更好的病变应该是侵袭性黑色素瘤。每个系统都应该由一个训练集开发，并由一个独立的测试集控制。每个系统都应由拥有最终决定权和责任的用户控制，并由独立的用户进行测试，不存在任何财务利益冲突。
Between 1987 and 2007, different groups developed digital image analysis systems for the diagnosis of benign and malignant skin tumors. As the result of significant differences in the technical devices, the number, the nature and benign/malignant ratio of included skin tumors, different variables and statistical methods any comparison of these different systems and their results is difficult. For the use and comparison of the diagnostic performance of different digital image analysis systems in the future, some principle basic conditions are required: All used systems should have a standardized recording system and calibration. First, melanocytic and nonmelanocytic lesions should be included for the development of the diagnostic algorithms. Critical analyses of the results should answer the question if in future only melanocytic lesions should be analyzed or all pigmented and nonpigmented lesions. This will also lead to the answer if only dermatologists or all specialities of medical doctors will use such a system. All artifacts (eg, hairs, air bubbles) should be removed. The number of variables should be chosen according to the number of included melanomas. A high number of benign skin lesions should be included. Of all lesions only 10% or better less should be invasive melanomas. Each system should be developed by a training-set and controlled by an independent test-set. Each system should be controlled by the user with the final decision and responsibility and tested by independent users without any conflict of financial interest.