NIR-SC-UFES: A portable NIR spectral dataset to skin cancer in vivo

In recent advancements, computer-aided diagnostic (CAD) for skin lesions using images and metadata has grown, yet it lacks details on lesion molecular structure. NIR spectroscopy offers unseen insights for CAD. Given skin cancer's severity and the need for early detection, the absence of public datasets hampers machine and deep learning (MDL) applications in spectroscopy. Addressing this gap, UFES's Dermatology Assistance Program developed the NIR-SC-UFES dataset, capturing 714 NIR spectra spanning benign conditions and skin cancers. This resource aims to empower researchers in AI and Chemometrics for enhanced automated skin cancer diagnosis.

近年来，基于图像与元数据的皮肤病变计算机辅助诊断（Computer-Aided Diagnostic, CAD）技术领域已取得显著进展，但此类技术尚未能提供病变分子结构层面的细节信息。近红外（Near Infrared, NIR）光谱技术可为CAD提供此前未被发掘的关键洞察视角。鉴于皮肤癌的危害性与早期筛查的迫切需求，公开数据集的缺失阻碍了机器学习与深度学习（Machine and Deep Learning, MDL）在光谱学领域的应用落地。为填补这一研究空白，UFES皮肤病辅助项目组构建了NIR-SC-UFES数据集，该数据集共收录714条近红外光谱样本，涵盖良性皮肤病变与皮肤癌两类病例。本数据集旨在赋能人工智能与化学计量学领域的研究人员，助力实现更精准的自动化皮肤癌诊断。