Hyperspectral Wide Gap Second Derivative Analysis For In Vivo Detection Of Cervical Intraepithelial Neoplasia

Hyperspectral imaging technique, with its advantages of high accuracy and non-invasive, has potential to improve the present situation of in vivo detection of cervical precancer. In recent years, the application of hyperspectral imaging for the cervical cancer detection has attracted many national and international researches to study it. This research studied the hyperspectral reflectance imaging technique for in vivo detection of high-grade cervical intraepithelial neoplasia (CIN). We applied the hyperspectral imaging system to clinical trial and imaged a wide-field of entire cervical tissue. Then we used a spectral analysis called wide gap second derivative spectral to process and analyze the cervical reflectance spectrum. We also used a minimum distance (MD) image segmentation algorithm to successfully classify the entire cervical tissue into the three categories of normal, inflammation, and high-grade CIN. The main researches we had completed are as follows:(1) We acquired the hyperspectral data cube in a wavelength range from600nm to800nm, with an interval of2nm. In the method of wide-field imaging, non-uniform illumination has become an inevitable problem which has interfered the researchers to detect and observe the spectral characteristic of cervical tissue. Therefore, we had proposed a wide gap second derivative spectral analysis to process the reflectance spectrum we acquired form cervical tissue through our hyperspectral imaging system. Derivative spectroscopy is less sensitivity to illumination variations than other analysis methods. It can optimize the background noise effectively, so it can matches the spectral feature better and improve the data separability.(2) The hyperspectral imaging system is not only the bulky and expensive, the process of imaging is time-consuming. All these will hinder the promotion of this instrument. Furthermore, the time-consuming imaging will influence the image quality, for example, the relative position between camera and tissue changes. Hence, we would like to acquire the effective spectral characteristic to reduce the wavelengths we used if possible. The wide gap second derivative spectral analysis we proposed in this study has the potential to achieve it. The calculation of second derivative utilizes only three specific wavelengths with specific wavelength interval. Therefore, we studied to acquire an optimal second derivative value, the value has the ability to classify the entire cervical tissue into the three categories of normal, inflammation, and high-grade CIN, and the classification result is coincident with the biopsy pathologic result. Base on the method of second derivative spectral analysis, the wavelengths we used in this study range from600nm to800nm can form numerous spectral configurations with different wavelength interval and corresponding three wavelengths. Thus we had designed a screening filter based on the superior capability of tissue classification to acquire the best spectral configuration. The selected wavelength of620nm,696nm, and772nm with a specific wide gap76nm emerged as the optimal selection which can classify the cervical tissue into three categories rapidly and accurately.(3) After the wavelengths combinations were screened by the filter proposed above, the used wavelengths from600nm to800nm were narrowed to several specific wavelengths (620nm,696nm and772nm respectively with specific wavelength interval76nm) for classification. Then we take the selected wavelengths with a minimum distance (MD) segmentation algorithm based on the Euclidean distance, which classify the cervical tissue into the three categories of normal, inflammation, and high-grade CIN. The classification result was compared with the biopsy pathologic result and examined by experienced gynecology oncologist.(4) We explained and analyzed that how the selected wavelength (620nm,696nm and772nm respectively with specific wavelength interval76nm) effect on the cervical tissue reflectance spectrum. We had explored and review a large number of relevant literature to further study how the reflectance spectrum of cervical tissue change with the dysplasia development among the wavelength range from600nm to800nm. Based on the detailed studies about cervical tissue structure and how the concentration of its components change, combined the selected specific wavelengths, we analyzed that how all these changes impact on cervical spectral response.(5) We had also studied the optical modeling of cervical tissue in the wavelength range from350nm to650nm to reproduce cervical reflectance spectrum rapidly and accurately. Modeling study is necessary to understand in depth how the presence of dysplasia changes the optical properties of cervical tissue, and to provide a quantitative understanding of the specific contributions of physiological dysplastic changes to the overall spectral response.