Wu Di, Fedorov Kukk Anatoly, Panzer Rüdiger, Emmert Steffen, Roth Bernhard
Hannover Centre for Optical Technologies, Leibniz University Hannover, Hanover, Germany.
University Medical Center Rostock, Rostock, Germany.
J Biophotonics. 2025 Sep;18(9):e70040. doi: 10.1002/jbio.70040. Epub 2025 Apr 21.
A multimodal method comprising optical imaging using OCT and molecular detection using Raman spectroscopy was developed to explore its capability for noninvasive differentiation between melanoma skin cancer and benign skin lesions. Key OCT parameters like the attenuation coefficient, R , and RMSE, extracted through exponential fitting, were incorporated into machine learning, achieving 96.9% accuracy and an AUC-ROC of 0.99 in 10-fold cross-validation. Raman spectroscopy revealed differences in carotenoid, amide-I, and CH-CH structures between melanoma and nevi, supporting the OCT findings. Autofluorescence background intensity variations further distinguished lesion types and enhanced lesion assessment. Future work will include the investigation of larger patient groups and the combination of both data sets in a combined algorithm. Also, the integration of both modalities and the developed method with photoacoustic tomography and high-frequency ultrasound appears beneficial toward achieving an optical biopsy of melanoma skin cancer and improving diagnostics.
开发了一种多模态方法，包括使用光学相干断层扫描（OCT）进行光学成像和使用拉曼光谱进行分子检测，以探索其对黑色素瘤皮肤癌和良性皮肤病变进行无创区分的能力。通过指数拟合提取的关键OCT参数，如衰减系数、R和均方根误差（RMSE），被纳入机器学习，在10折交叉验证中实现了96.9%的准确率和0.99的曲线下面积（AUC-ROC）。拉曼光谱揭示了黑色素瘤和痣之间类胡萝卜素、酰胺-I和CH-CH结构的差异，支持了OCT的研究结果。自体荧光背景强度变化进一步区分了病变类型并增强了病变评估。未来的工作将包括对更大患者群体的研究，以及在组合算法中对两个数据集进行组合。此外，将这两种模式以及所开发的方法与光声断层扫描和高频超声相结合，似乎有利于实现黑色素瘤皮肤癌的光学活检并改善诊断。
