高光谱技术在皮肤检测中的实现：构建高效系统与鲁棒模型

Implementation of Hyperspectral Technology in Skin Detection: Building Efficient Systems and Robust Models

在上一篇文章中，我们探讨了高光谱成像技术在皮肤检测中的潜力，而本文将关注如何实现这一技术的实现。

In the previous article, we explored the potential of hyperspectral imaging technology in skin detection. This article will focus on its practical implementation.

皮肤样品的可见光和近红外光谱 / Visible and Near-Infrared Spectra of Skin Samples

为实现高光谱成像技术的有效应用，多个研究团队搭建了各具特色的高光谱成像系统。其中一个西班牙团队，搭建了不同的系统。他们使用398.08~995.20nm的高光谱相机，配备了电动底座和卤素光源，以优化成像质量，确保稳定的数据采集。

该团队还搭建了，采用900~1700nm的光谱范围，搭建系统时特别关注患者的舒适度，设计了支撑装置，让患者在拍摄过程中能够稳定休息。此装置由金属梁和多个3D打印支撑平台构成，提供了柔软且适应不同部位的支持。

To effectively apply hyperspectral imaging, multiple research teams have developed specialized systems. One Spanish team, for instance, constructed distinct setups. They employed a hyperspectral camera covering 398.08–995.20 nm, equipped with a motorized stage and halogen lighting to optimize imaging quality and ensure stable data acquisition.

The team also developed another system operating in the 900–1700 nm range, prioritizing patient comfort by incorporating a support device that allowed subjects to remain stable during imaging. This setup consisted of metal beams and multiple 3D-printed support platforms, providing soft and adaptable positioning for different body areas.

可见光系统 / the visible light system

近红外系统。（a）本研究中为数据采集目的而构建的高光谱推扫平台。（b）在采集过程中帮助患者感到舒适的不同支持平台。

the near-infrared system. (a) The hyperspectral push-broom platform constructed for data collection in this study. (b) Various support platforms designed to enhance patient comfort during acquisition.

在数据分析方法上，近年来的研究主要集中在机器学习模型的应用。传统的简单图像处理方法虽然实现直接，但在应对复杂皮肤病变时，其效果往往不能令人满意。机器学习模型，为皮肤检测的准确性提供了支持，这些模型具备良好的泛化能力，能够在多种条件下有效识别不同类型的皮肤病变。

Recent research has increasingly focused on machine learning models for data analysis. While traditional image processing methods are straightforward, their performance in detecting complex skin lesions is often unsatisfactory. Machine learning models, however, offer superior accuracy and generalization, enabling reliable identification of diverse skin lesions under varying conditions.

在一个研究中，科研人员对不同的分类和分割方法进行了比较。这些方法各有优缺点，支持向量机在高维空间中表现良好，随机森林对过拟合有一定的鲁棒性，K均值聚类适用于简单的分类任务，而主成分分析则有效进行降维，保留数据中的重要特征。这具体取决于组织和目标病变的类型。

In one study, researchers compared different classification and segmentation approaches, each with unique strengths:

·Support Vector Machines (SVM) excel in high-dimensional spaces.

·Random Forests demonstrate robustness against overfitting.

·K-means Clustering is suitable for simpler classification tasks.

·Principal Component Analysis (PCA) effectively reduces dimensionality while preserving critical features.

·The optimal method depends on tissue type and the target lesion.

各类方法的比较（部分）/ Comparison of different methodologies (partial)

前面提到的西班牙团队，该团队利用近红外高光谱成像技术，针对基底细胞癌（BCC）和皮肤鳞状细胞癌（SCC）进行了检测，强调使用鲁棒特征统计方法来进行数据分析。该方法不仅提高了系统的稳定性，还确保在样本中存在噪声和异常值时，依旧能获得较高的检测准确性。

the aforementioned Spanish team utilized near-infrared hyperspectral imaging to detect basal cell carcinoma (BCC) and squamous cell carcinoma (SCC), emphasizing robust statistical feature extraction. This approach not only improved system stability but also maintained high detection accuracy despite noise and outliers.

上图：使用每个样本在各个波长上的中位数值所得到的鲁棒特征。

下图：使用平方根的双权重中方差作为每个样本变异度的测量方法，得到这些样本的鲁棒偏差。

Top: Robust features derived from median values of each sample across wavelengths.

Bottom: Robust deviations calculated using the square root of the biweight midvariance (√BWMV) as a measure of variability.

此外，他们在另一个实验中重点关注BCC、SCC和AK（光化性角化病）与健康皮肤的差异，同样采用了鲁棒统计方法，同时还使用多变量统计分析进行样本间的比较，以发现数据中潜在的差异。

In another experiment, the team examined differences among BCC, SCC, actinic keratosis (AK), and healthy skin, again applying robust statistics alongside multi-variate analysis to uncover subtle data variations.

在本研究中通过多种方法确定的最佳定义窗口。虚线垂直线所划定的区域标记了573.45nm至779.88nm之间最终感兴趣的窗口。

Optimal spectral window (573.45–779.88 nm, marked by dashed vertical lines) identified through multiple methods in this study.

每个样本的高光谱特征。（a）鲁棒特征标记了中央倾向以及5%和95%百分位置信区间（下线和上线分别）。（b）√BWMV计算表示鲁棒样本方差。

Hyperspectral features of each sample. (a) Robust features indicating central tendency with 5% and 95% percentile confidence intervals (lower and upper bounds, respectively). (b) √BWMV representing robust sample variance.

综上所述，高光谱成像技术在皮肤检测中展现出了优势，尤其是在系统构建与模型泛化能力方面。通过选择适宜的波长范围，结合先进的数据分析技术，我们的高光谱相机在皮肤疾病早期检测中提供了坚实的基础。

值得一提的是，我们公司不仅销售高光谱相机，还能提供专业的硬件技术支持，助力您的研究与应用提升效率。未来，随着高光谱成像技术与机器学习的深度融合，该领域必将迎来更多机会，相信皮肤癌的早期检测将变得更加高效和可靠，为患者带来更大的福音。

Hyperspectral imaging demonstrates unique advantages in skin detection, particularly in system design and model generalization. By selecting optimal wavelength ranges and integrating advanced analytics, hyperspectral cameras provide a robust foundation for early skin disease diagnosis.

Notably, our company not only supplies hyperspectral cameras but also offers expert hardware support to enhance research and application efficiency. As hyperspectral imaging and machine learning continue to converge, this field holds immense promise—ushering in more efficient, reliable early detection of skin cancer and greater benefits for patients.

案例来源 / Source：

1. Courtenay LA, González-Aguilera D, Lagüela S, Del Pozo S, Ruiz-Mendez C, Barbero-García I, Román-Curto C, Cañueto J, Santos-Durán C, Cardeñoso-Álvarez ME, Roncero-Riesco M, Hernandez-Lopez D, Guerrero-Sevilla D, Rodríguez-Gonzalvez P. Hyperspectral imaging and robust statistics in non-melanoma skin cancer analysis. Biomed Opt Express. 2021 Jul 20;12(8):5107-27. doi: 10.1364/BOE.428143. PMID: 34513245; PMCID: PMC8407807.

2. Courtenay LA, Barbero-García I, Martínez-Lastras S, Del Pozo S, Corral de la Calle M, Garrido A, Guerrero-Sevilla D, Hernandez-Lopez D, González-Aguilera D. Near-infrared hyperspectral imaging and robust statistics for in vivo non-melanoma skin cancer and actinic keratosis characterisation. PLoS One. 2024 Apr 25;19(4):e0300400. doi: 10.1371/journal.pone.0300400. PMID: 38662718; PMCID: PMC11045066.

3. Aloupogianni E, Ishikawa M, Kobayashi N, Obi T. Hyperspectral and multispectral image processing for gross-level tumor detection in skin lesions: a systematic review. J Biomed Opt. 2022 Jun 8;27(6):060901. doi: 10.1117/1.JBO.27.6.060901.