Abstract:
Optical spectroscopy is studied to contribute to skin cancer diagnosis. Indeed, optical spectra are modified along cancer progression and provide complementary information (e.g., on metabolism and tissue structure) to clinical examination for surgical guidance [1,2]. The current original dataset is made of autofluorescence and diffuse reflectance spectra acquired in vivo on 131 patients\' skin with the SpectroLive device [3,4]. Spatially-resolved spectroscopy measurements were performed using a multi-fiber optic probe featuring 4 distances (0.4-1 mm) between excitation and collection optical fibers: spatial resolution allows spectra acquired at different distances to carry information from different depths in skin tissues. Five types of autofluorescence spectra were acquired using five different wavelength excitations (on the 365-415 nm spectral range) in order to collect information on several skin endogenous fluorophores (e.g., flavins, collagen). A sixth light source (white broadband) was used to acquire diffuse reflectance spectra carrying information about skin scattering properties and skin endogenous absorbers such as melanin and hemoglobin. Patients were proposed to be included into the clinical trial if they were suspected of suffering from actinic keratoses (precancerous skin lesions) or from basal or squamous cell carcinomas: in all cases, complete diagnostics is provided in the dataset. To increase the interest of the dataset and evaluate the dependence of optical spectra (intensity, shape) not only on pathological states but also on healthy skin features (civil age, skin age, gender, phototype, anatomical site), spectra were acquired for all 131 patients on two so-called \"reference\" skin sites known to rarely suffer from skin cancer: palm of the hand (featuring a thick skin type) and inner wrist (featuring thin skin). Spectra are available in .tab files: first column displays the spectral range on which intensity spectra were recorded (317-788 nm) and each following column provides an intensity spectrum acquired by each spectrometer for a given combination of light source excitation and distance. Each of the 131 folders corresponding to each of the 131 patients contains a .json file providing patients clinical features: gender, civil age, skin age, phototype score and class. All .tab files names include anatomical site and anatomopathological diagnostics of the skin site on which spectra were acquired: codes were defined to match a letter or an acronym to each diagnostic and anatomical site. To ensure quality control, a spectrum was acquired on the same calibration standard before starting spectra acquisition on each patient. It is therefore possible to follow the impact of the acquisition optical chain ageing during the 4.5 years that the patients were included. This dataset can be used by epidemiologists for the characterization of populations affected by skin cancers (gender ratio, mean age, anatomical sites typically affected, etc.); it may also be used by researchers in artificial intelligence to develop innovative methods to process such data and contribute to non-invasive diagnostics of skin cancers whose incidence is steadily increasing.
摘要:
研究了光谱学有助于皮肤癌诊断。的确,光谱随着癌症进展而改变,并提供补充信息(例如,在代谢和组织结构上)以临床检查为手术指导[1,2]。当前的原始数据集是由使用SpectroLive设备[3,4]在131名患者皮肤上体内采集的自发荧光和漫反射光谱组成。使用具有激发和收集光纤之间的4个距离(0.4-1mm)的多光纤探头进行空间分辨光谱测量:空间分辨率允许在不同距离处获取的光谱携带来自皮肤组织中不同深度的信息。使用五种不同波长激发(在365-415nm光谱范围内)采集了五种类型的自发荧光光谱,以收集有关几种皮肤内源性荧光团的信息(例如,黄素,胶原蛋白)。第六光源(白色宽带)用于获取携带关于皮肤散射性质和皮肤内源性吸收剂(例如黑色素和血红蛋白)的信息的漫反射光谱。如果患者被怀疑患有光化性角化病(癌前皮肤病变)或基底或鳞状细胞癌,则建议将其纳入临床试验:在所有情况下,数据集中提供完整的诊断。为了增加数据集的兴趣并评估光谱的依赖性(强度,形状)不仅在病理状态上,而且在健康的皮肤特征上(民用年龄,皮肤年龄,性别,照型,解剖部位),光谱是在两个所谓的“参考”皮肤部位获得的,已知很少患有皮肤癌:手掌(具有厚皮肤类型)和内手腕(具有薄皮肤)。光谱可在。选项卡文件:第一列显示记录强度光谱的光谱范围(317-788nm),并且下一列中的每一列提供由每个光谱仪针对光源激发和距离的给定组合获得的强度光谱。与131名患者中的每一个对应的131个文件夹中的每一个包含a.JSON文件提供患者临床特征:性别,民事年龄,皮肤年龄,照型分数和类别。全部。选项卡文件名包括采集光谱的皮肤部位的解剖部位和解剖病理学诊断:定义代码以匹配字母或缩写到每个诊断和解剖部位。为了确保质量控制,在开始对每位患者进行光谱采集之前,在相同的校准标准品上采集光谱。因此,可以跟踪在包括患者的4.5年期间采集光链老化的影响。流行病学家可以使用此数据集来表征受皮肤癌影响的人群(性别比例,平均年龄,通常受影响的解剖部位,等。);人工智能领域的研究人员也可能使用它来开发创新的方法来处理此类数据,并有助于对发病率稳步上升的皮肤癌进行非侵入性诊断。