There are many types of skin lesions which can be classified into three main categories of malignant, pre-malignant and benign. Malignant lesions are cancerous, benign lesions are harmless and pre-malignant lesions may turn into skin cancer.
Histopathological analysis, which is defined as the microscopic studies of tissues from the skin, remains the gold standard to determine the classification of skin lesions. Near-IR
Spectroscopy is an in-vivo method which measures the reflection spectra from different types of skin lesions and after preprocessing the spectra could be subjected to a
Linear Discriminant Analysis (LDA) to classify different types of skin lesions. In one study (Reference 1) the wavelength range from 400-2500 nm was incident on skin lesions.
The following set-up below shows the reflection probe whose two inputs are connected to a Halogen source and a Near-IR spectrometer and its output shines the Near-IR
beam onto the skin sample and then collects the diffused reflection which carries the important information about the skin condition towards the spectrometer.
Figure 1 below shows the setup.
Figure 1: Reflection probe setup for skin cancer detection using NIR spectroscopy (Reference 2)
Skin cancer is the most malignant and common condition as far as skin diseases are concerned. A correct and timely diagnosis of skin cancer is crucial in treatment of this disease. The current diagnosis tool for skin cancer involves three steps. In step 1, the physician asks the patients questions about the history of the disease. In step 2, there will be a visual examination of the lesion and in step 3 a biopsy is done which is examined under the microscope. Near-IR spectroscopy can be used as a complementary non-invasive method to further confirm the diagnosis.
There are several types of malignant, pre-malignant, and benign lesions. The two most dangerous types of cancers are called Basal Cell Carcinoma (BCC) and malignant
melanoma. There are also other types of harmless skin conditions such as Benign Nevus and Solar lentigo. A few of these skin lesions are shown in Figure 2.
Figure 2: Different types of malignant, premalignant and benign skin lesions (Reference 1)
The challenges in diagnosis of skin cancers involve differentiation between malignant and pre-malignant lesions, malignant and benign lesions, pre-malignant and benign lesions and also identification of different types of cancer. After the cancer type has been diagnosed, there are challenges related to treatment, identification of depth of cancer in the skin, identifying the borders of the area of cancer on the skin, etc. Near-IR spectroscopy can help in all of these tasks. Near-IR beam is mostly scattered than absorbed in the skin which makes it nearly transparent to the skin. The idea of using Near-IR spectroscopy is to measure the reflectance spectra from the skin, convert it to absorbance pre-processing the data, do a LDA plot on the absorbance spectra to differentiate between different types of cancer.
In one study, nearly 200 reflection spectra were collected from nearly 150 patients where the dermatologist had ordered a biopsy (Reference 1). The wavelength range for the measurement was 400-2500 nm which encompassed both visible and Near-IR spectra and each measurement took about 40 seconds. The active area of the probe as shown in Figure 1 was 7 mm diameter and the lesion was kept at 1 mm distance from the tip of the probe. More details about the details of the experimental setup can be found in Reference 1. After the acquisition of the spectra a biopsy was performed on the lesion which classified them into 6 categories as follows:
- Actinic keratosis (pre-malignant)
- Basal cell carcinoma (malignant)
- Actinic lentigo (benign)
- Dysplastic nevi (pre-malignant)
- Benign nevi (benign)
- Seborrheic keratosis (benign)
The pre-processing of the spectra selected the region 400-1840 nm for analysis with each spectrum containing 720 data points. After performing Standard Normal Variate scatter correction and averaging of spectra from similar lesions, fewer spectra were left. These were 33 actinic keratoses, 34 basal cell carcinoma, 13 dysplastic novi, 12 actinic lentigines, 22 benign nevi and 18 seborrheic keratoses. The spectra of normal skin were also recorded under similar conditions and settings and were subtracted from the lesion spectra to magnify the spectral differences between the lesions. LDA analysis was performed on the original spectra and the subtracted spectra were only plotted to observe the differences between lesions. Both of these plots are shown in Figure 3.
Figure 3: Absorbance spectra on the left and lesion minus normal spectra on the right (Reference 1)
As far as subtracted spectra were concerned, the notably different subtracted spectrum was that of dysplastic nevi. The other difference spectra were more or less the same. A 2D LDA analysis was performed on the absorbance data. An accuracy of 71% resulted from distinguishing malignant from pre-malignant lesions. Also, 77% accuracy to distinguish between pre-malignant from benign and 85% in distinguishing malignant from benign. For comparison’s sake, the clinical accuracies in distinguishing the three pairs mentioned were 90, 87 and 91.5% respectively. The near overall accuracy of 72-98% in identifying different types of cancer.
There were also other studies which used different wavelength ranges and processing methods. In a recent study (Reference 3, 2024) a wavelength range of 900-1700 nm was used only and the machine learning methods such as XGBoost, CatBoost. LightGBM and 1D-convolutional neural networks were used. The best performance was obtained by LightGBM with 84% accuracy and 85% precision. In another study mid-IR wavelength range and FTIR was used for skin cancer diagnosis (Reference 4).
ASP Laser Inc offers Nirvascan spectrometers in different spectral ranges such as 900-1700 nm, 1350-2150 and 1600-2400 nm. These spectrometers are cost-friendly and accurate (10 nm spectra resolution, +/- 0.7 nm accuracy). ASP Laser Inc also provides the reflection probe and the halogen source for a complete setup.
Please check the following link for more information on the 900-1700 nm Nirvascan fiber optic model.
Smart Low Cost Fiber Spectrometer Shortwave Infrared 900-1700nm
References:
- Near-infrared spectroscopy for dermatological applications, L.M. McIntosh, et.al, vibrational spectroscopy 28, 2002
- Investigating the influence of probe pressure on human skin using diffuse reflection spectroscopy, I Ahmed, et.al, Micromachines 2023 14(10)
- Skin cancer diagnostics using NIR spectroscopy data of skin lesions in vivo using machine learning algorithms, M. Rocha, et.al, Biocybernectics and biomedical engineering 44 (2024)
- FT-IR spectroscopy study in early diagnosis of skin cancer, M Kyriyakidou, et.al, in vivo 31: 1131-1137 (2017)
Author: Rez Mani (PhD)