Surface-enhanced Raman spectroscopy + support vector machine: a new noninvasive method for prostate cancer screening?

Abstract

Prostate cancer is one of the most common malignancies of the older males worldwide. Early diagnosis and treatment are important to improve the survival of patients. Recently, we developed a new method for prostate cancer screening: by measuring the serum surface-enhanced Raman spectroscopy of prostate cancer patients and normal subjects, combining with classification algorithms of support vector machines, the measured surface-enhanced Raman spectroscopy spectra are successfully classified with accuracy of 98.1%. Although the practical application faces several difficulties, we believe that this label-free serum surface-enhanced Raman spectroscopy analysis technique combined with support vector machine diagnostic algorithms will become a powerful tool for noninvasive prostate cancer screening in the future.

Keywords:

• classification
• prostate cancer
• serum
• support vector machine
• surface enhanced Raman spectroscopy

Prostate cancer is one of the most common malignancies of older males worldwide and is the sixth leading cause of cancer-related death. Every year, there are about 899,000 new cases and 260,000 mortalities, accounting for 6% of all cancer deaths globally. About one in every six men will suffer from the disease during their lifetime [1]. It is important to improve the survival of patients by early diagnosis and treatment. Currently, there are many diagnostic methods including B-mode ultrasound, CT scan, biopsy and prostate specific antigen (PSA) screening, but all these techniques have various limitations. For example, B-mode ultrasound only discerns the solid tumor, which suggests that patients are not in early stage of cancer. Biopsy is the gold standard of cancer examination, but it is invasive and impractical for high-risk patients with multiple suspicious lesions. PSA detection is the important screening means, but the PSA test can lead to overdiagnosis because PSA levels are low in some subjects with prostate cancer, so the US Preventive Services Task Force does not recommend screening for prostate cancer based on PSA at present. Recently, the surface-enhanced Raman spectroscopy (SERS) in combination with support vector machine (SVM) has been successfully applied in classification of prostate cancer from normal subjects by detecting peripheral blood [2]. We feel that blood SERS will open a new approach for noninvasive screening prostate cancer at the early stages.

SERS is a special phenomenon of Raman scattering. When sample molecules are attached on the surface of metal nanostructure, the signal of Raman scattering can be enhanced to a great extent; this phenomenon is called SERS [3]. SERS can provide spectroscopic fingerprint-type information about the molecular composition and structures. When compared with other optical spectroscopy techniques, such as the fluorescence spectroscopy and infrared absorption spectroscopy, SERS holds many significant advantages. First of all, its sensitivity is so high that SERS has been successfully applied to probing a single molecule. Second, Raman spectrum exhibits sharp spectral features that are characteristic of specific molecular structures. Third, the measurement time of SERS is not limited, because there is no photobleaching in SERS [4]. Therefore, the SERS is very suitable for studying biological samples.

Peripheral blood samples are ideal materials for noninvasive diagnosis, which can be executed expediently and even repeatedly throughout the treatment duration of patients, and sampling and testing can be conducted continuously during the monitoring process of high-risk subjects. At the early stage of cancer, the contents and conformations of biomacromolecules, such as proteins, nucleic acid, lipids in blood will undergo subtle changes, which can be disclosed by SERS. Chen et al. conducted many pioneering researches and acquired fruitful outcome in this field with SERS [5–8].

Owing to tiny difference of blood SERS spectra between normal subjects and cancer patients, powerful data analysis algorithm must be adopted to draw effective information from several hundreds of spectral variables in measured SERS spectra. Several multivariate statistical algorithms, such as principal component analysis (PCA), linear discriminate analysis, artificial neural networks and SVM, have been used to develop diagnostic models [9,10]. Among these algorithms, SVM is considered to be superior over traditional linear methods because it has the ability to process binary classification problem with nonlinear boundary. The fundamental idea involves that SVM algorithm search for the optimal hyperplane that maximizes the margin of separation between the hyperplane and the closest data points on both sides of the hyperplane [11,12]. The main advantages are that it can efficiently classify the small samples, regardless of the distribution nature of samples.

In our study, we measured serum SERS spectra of patients with prostate cancer and normal volunteers. The huge increase of Raman peaks in many dominant vibration bands was observed. Three SVM classifier models, including linear, polynomial and radial basis functions kernel functions, were built and evaluated with measured serum SERS spectra. The maximum diagnostic accuracy of 98.1% was acquired with radial basis functions kernel SVM classifier model. This result is superior to the one obtained from PCA with the same dataset of serum SERS spectra. Other groups have also reported studies about blood SERS. For example, Feng et al. measured blood plasma SERS spectra of nasopharyngeal normal volunteers and cancer patients with silver nanoparticles. Combining PCA and linear discriminate analysis, they acquired the sensitivity of 90.7% and specificity of 100% for classifying patients with nasopharyngeal cancer [5]. Lin et al. explored variability of different tumor (T) stages in nasopharyngeal cancer with blood plasma SERS. The diagnostic accuracy of 83.5 and 93.3%, respectively, was achieved for classification between early T (T1) stage cancer and normal and advanced T (T2–T4) stage cancer and normal blood groups [8].

All the studies mentioned above confirm the fact that SERS in combination with multivariate statistical algorithms has the ability of classifying blood SERS spectra between normal subjects and cancer patients with high specificity and sensitivity. The facts suggest that there is great potential for noninvasive cancer screening with SERS through peripheral blood samples. Although these studies have yielded encouraging results, yet there is long way to go before the method is applied clinically. First, the patients who provided blood samples for our study were almost in the middle or advanced stages of cancer (70% in high grade), so the diagnostic models built with these serum samples SERS spectra belong to corresponding cancer staging. Whether these models can be applied to diagnose early cancer is unknown. However, if the samples of early cancer patients are increased and the corresponding diagnostic models are constructed, we think this problem can be resolved easily. Second, the classification of serum samples in experiments is confined between normal volunteers and single type of cancer patients. Whether the studied cancer patients can be differentiated from other types of cancer or disease patients is unclear. The problem needs to be further researched carefully. Previously, we had explored the possibility of discriminating cancer patients and normal subjects with serum SERS and SVM through peripheral blood samples [13]. One hundred and thirty blood samples were obtained from patients with liver cancer, esophageal cancer, gastric cancer, nasopharyngeal cancer and colonic cancer. The results showed that the serum SERS spectra of cancer patients can be differentiated from normal subjects with sensitivity of 92.3% and specificity of 98.2%. The preliminary results inspired us to develop a new cancer screening scheme with SERS and SVM: patients with a positive result could be considered for further investigation; whereas a negative result would decrease the need for further expensive checks. Next, we will further study the differences of serum SERS spectra among different types of cancer patients and establish appropriate diagnostic models with SVM technique. We believe that, with the development of technique and continued research, these difficulties will eventually be resolved in the future.

In the course of applying SERS technique, it is important to adopt appropriate SERS-active nanoparticles because the reproducibility of the Raman scattering signals is related to their homogeneity, stability, biocompatibility and enhanced capability [14]. The silver nanoparticles enhance the Raman signals to a great extent, but the homogeneity is unsteady. In contrast, gold nanoparticles have excellent homogeneity, but their ability to enhance Raman signals is not as good as that of silver particles. Recently, Tian group developed a new approach named shell-isolated nanoparticle-enhanced Raman spectroscopy, in which the Raman signal amplification is provided by gold nanoparticles with an ultra-thin silica or alumina shell. We think that this kind of new SERS-active nanoparticles is beneficial to our study because of its excellent stability, homogeneity, biocompatibility and enhanced capability [15].

SVM is a powerful tool for classifying data with complex boundary. The basic principle is that, for the nonlinear nonseparable sample sets, SVM maps the sample data to a higher dimensional feature space to linearize the boundary of sample sets by specific kernel functions. Our results show that the classification accuracy of SVM is higher than that of PCA, suggesting the existence of nonlinear boundary for serum SERS datasets between normal subjects and cancer patients. Although SVM has a strong ability of classification, it has its own limitations in research. The basis of classification is the differences of serum SERS spectra between normal subjects and cancer patients. However, it does not know what differences and corresponding characteristic spectra produce key contributions to the classification of SVM. The information is important to identify and understand the molecular mechanisms of cancer. PCA technique also does not overcome the same difficulty. Perhaps it will be helpful to resolve the problem with the aid of feature selection techniques, such as genetic algorithms. Huang’s group and our group have conducted similar studies by feature selection [16–18].

In conclusion, blood SERS combined with SVM could classify prostate cancer and normal subject with high accuracy. Although there are some difficulties with practical applications, we believe that, with the development of technique and continued research, this label-free serum SERS analysis technique combined with SVM diagnostic algorithm will become a powerful tool for noninvasive prostate cancer screening in the future.

Acknowledgements

The authors would like to acknowledge the financial support of the Medical Research Foundation of Guangdong Province (A2014466), Doctor Start Fund of Guangdong Medical College (XB1407), the National Natural Science Foundation of China (61335011, 61275187 and 31300691), Specialized Research Fund for the Doctoral Program of Higher Education of China (20114407110001), the Natural Science Foundation of Guangdong Province (9251063101000009) and the Cooperation Project in Industry, Education and Research of Guangdong province and Ministry of Education of China (2011A090200011).

Financial & competing interests disclosure

The authors have no relevant affiliations or financial involvement with any organization or entity with a financial interest in or financial conflict with the subject matter or materials discussed in the manuscript. This includes employment, consultancies, honoraria, stock ownership or options, expert testimony, grants or patents received or pending, or royalties.

No writing assistance was utilized in the production of this manuscript.