Loading...

Hyperspectral Imaging Combined with Chemometric Techniques for Diagnosis of Breast Cancer

Roshandel, Pegah | 2023

117 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 55902 (06)
  4. University: Sharif University of Technology
  5. Department: Chemistry
  6. Advisor(s): Parastar Shahri, Hadi
  7. Abstract:
  8. Breast cancer is one of the most known types of cancer. About every eight women, one woman will suffer from one of the types of malignant tissues during her life. Diagnosing this type of cancer in the early stages is an important matter and can lead to full recovery. Therefore, one of the challenges in this field is the emergence of a fast method with high sensitivity to diagnose this disease in its early stages. Currently, biopsy is the standard method for breast cancer diagnosis. However, there are some drawbacks to this method. For instance, in order to detect the tumor margin, all breast tissue must be removed, which causes all breast tissue, including healthy tissues, to be removed. Hyperspectral imaging is an emerging technique in the field of medicine and has provided promising results in the field of cancer diagnosis. Hyperspectral imaging is a type of spectral imaging that provides useful information regarding the spatial distribution of species and the nature of existing species. In order to extract useful and desired information from these 3D images, powerful methods of multivariate analysis or chemometrics, which are capable of analyzing high-order data, are needed. The aim of the current project is to develop different multivariate classification methods based on the approach of selecting important pixels as well as multivariate curve resolution altering least squares (MCR-ALS) in order to obtain pure spatial and spectral profiles for species in hyperspectral images ranging from 400 to 950 nm. According to our studies, no such study has been done with the mentioned algorithm for cancer diagnosis. Thus, four important approaches for distinguishing two-class and three-class classifications were carried out in this research. In the first approach, the data obtained from the hyperspectral imager were analyzed without further data preprocessing with PCA and PLS-DA methods in three and two classes. The accuracy of binary and three class classifications, in the prediction set, is equal to 70.1% and 42.4%, respectively. The results indicate good accuracy for constructing a diagnostic model, but its undeniable drawback is the time-consuming implementation. For this reason, the overall average of each sample was used to apply PCA and PLS-DA methods. The accuracies obtained in the two and three-class methods, in the prediction set were 63.0% and 45.4%, respectively, which shows that important information was lost. According to this reason, in the next step, the binning method was used, and then the abovementioned methods were used again. The accuracies obtained for two-class classification in the prediction set are 59.7% and for three-class 51.2%. which shows that the performance of this method is similar to the raw data. In the next approach, the spatial profile of the main principal component was obtained from the MCR-ALS algorithm, and the PCA and PLS-DA were implemented on the spatial profile of the main component. The accuracy results of PLS-DA for binary and three-class models in the prediction set were 81.1% and 45.4%, respectively, which are the best results so far. For this reason, we use this type of data in other methods, for example, DD-SIMCA, which is a soft method, as can be seen in the third chapter. Also, a non-linear method such as SVM was used, which showed an accuracy equal to 50%. This result is not as good as the previous method and is not able to classify three different classes. We conclude that by using the MCR-ALS method and then PLS-DA, it is possible to differentiate the type of cancerous tissue, non-cancerous tissue, and even the degree of cancer. The results of MCR-ALS provide the possibility to differentiate between pixels of healthy tissue and cancerous tissue.
  9. Keywords:
  10. Breast Cancer ; Multivariate Data Analysis ; Hyperspectral Images ; Chemometrics Method ; Cancer Diagnosis

 Digital Object List

 Bookmark

No TOC