JAMA Dermatology ( IF 10.9 ) Pub Date : 2020-01-01 , DOI: 10.1001/jamadermatol.2019.3360 Philipp Tschandl 1
In recent years we have seen experimental evidence suggesting that neural networks are able to detect skin cancer via dermatoscopic and clinical images.1-4 Most studies used preprocessed and cropped images, in which parts of the background have been removed to fit to a predefined size. Custom preparation of images, however, is not feasible for laypersons or nonspecialists. In this issue of JAMA Dermatology, Han et al5 are opening the stage of automated skin cancer recognition without the need for this type of preprocessing. They applied a technique called object detection on a large data set including a variety of skin diseases.
中文翻译:
用于皮肤癌识别的自动目标检测的问题和潜力。
近年来,我们已经看到实验证据表明神经网络能够通过皮肤镜和临床图像检测皮肤癌。1 -4大多数研究都使用预处理和裁剪后的图像,其中背景的一部分已被删除以适合预定义的大小。但是,对于外行人员或非专业人员而言,定制图像准备是不可行的。在本期《JAMA皮肤病学》中,Han等人5开启了自动皮肤癌识别的阶段,而无需进行此类预处理。他们在包括各种皮肤疾病的大型数据集上应用了一种称为对象检测的技术。