Hoffmann K, Gambichler T, Rick A, Kreutz M, Anschuetz M, Grünendick T, Orlikov A, Gehlen S, Perotti R, Andreassi L, Newton Bishop J, Césarini J-P, Fischer T, Frosch P J, Lindskov R, Mackie R, Nashan D, Sommer A, Neumann M, Ortonne J P, Bahadoran P, Penas P F, Zoras U, Altmeyer P
Department of Dermatology, Ruhr-University Bochum, Gudrunstrasse 56, D-44791 Bochum, Germany.
Br J Dermatol. 2003 Oct;149(4):801-9. doi: 10.1046/j.1365-2133.2003.05547.x.
Early detection of melanomas by means of diverse screening campaigns is an important step towards a reduction in mortality. Computer-aided analysis of digital images obtained by dermoscopy has been reported to be an accurate, practical and time-saving tool for the evaluation of pigmented skin lesions (PSLs). A prototype for the computer-aided diagnosis of PSLs using artificial neural networks (NNs) has recently been developed: diagnostic and neural analysis of skin cancer (DANAOS).
To demonstrate the accuracy of PSL diagnosis by the DANAOS expert system, a multicentre study on a diverse multinational population was conducted.
A calibrated camera system was developed and used to collect images of PSLs in a multicentre study in 13 dermatology centres in nine European countries. The dataset was used to train an NN expert system for the computer-aided diagnosis of melanoma. We analysed different aspects of the data collection and its influence on the performance of the expert system. The NN expert system was trained with a dataset of 2218 dermoscopic images of PSLs.
The resulting expert system showed a performance similar to that of dermatologists as published in the literature. The performance depended on the size and quality of the database and its selection.
The need for a large database, the usefulness of multicentre data collection, as well as the benefit of a representative collection of cases from clinical practice, were demonstrated in this trial. Images that were difficult to classify using the NN expert system were not identical to those found difficult to classify by clinicians. We suggest therefore that the combination of clinician and computer may potentially increase the accuracy of PSL diagnosis. This may result in improved detection of melanoma and a reduction in unnecessary excisions.
通过各种筛查活动早期发现黑色素瘤是降低死亡率的重要一步。据报道，计算机辅助分析皮肤镜检查获得的数字图像是评估色素沉着性皮肤病变（PSL）的一种准确、实用且省时的工具。最近开发了一种使用人工神经网络（NN）对PSL进行计算机辅助诊断的原型：皮肤癌诊断与神经分析（DANAOS）。
为了证明DANAOS专家系统对PSL诊断的准确性，在不同的多国人群中进行了一项多中心研究。
开发了一种校准的摄像系统，并用于在9个欧洲国家的13个皮肤科中心进行的多中心研究中收集PSL的图像。该数据集用于训练用于黑色素瘤计算机辅助诊断的NN专家系统。我们分析了数据收集的不同方面及其对专家系统性能的影响。NN专家系统使用包含2218张PSL皮肤镜图像的数据集进行训练。
所得专家系统的表现与文献中发表的皮肤科医生的表现相似。性能取决于数据库的大小和质量及其选择。
本试验证明了对大型数据库的需求、多中心数据收集的有用性以及从临床实践中收集具有代表性病例的益处。使用NN专家系统难以分类的图像与临床医生发现难以分类的图像并不相同。因此，我们建议临床医生和计算机相结合可能会提高PSL诊断的准确性。这可能会改善黑色素瘤的检测并减少不必要的切除。
