University of Twente Student Theses
Tissue type classification and resection margin estimation of colorectal cancer using ultrasound, elastography, and diffuse reflectance spectroscopy
Egmond, C.A.W. van (2021) Tissue type classification and resection margin estimation of colorectal cancer using ultrasound, elastography, and diffuse reflectance spectroscopy.
|
PDF
5MB |
| Abstract: | Introduction: Colorectal cancer (CRC) ranks third in terms of incidence worldwide, and second in terms of mortality. An oncologic resection with negative resection margins is essential for overall survival and local recurrence. Achieving negative circumferential resection margins (CRM) in locally advanced rectal cancer is problematic through the effects of neoadjuvant radiotherapy and the technically challenging procedure. Although diffuse reflectance spectroscopy (DRS) can distinguish colorectal tumor tissue from healthy tissue, it is not able to distinguish tumor from the neoadjuvant effect called fibrosis. Furthermore, it is challenging to determine the resection margin with DRS. Ultrasound (US) B-mode may be of additional value since the healthy colorectal wall and fat are discriminative from tumor tissue. However, it is difficult to distinguish tumor tissue from fibrosis through the same ultrasonic appearance. Ultrasound elastography is of additional value to US B-mode in colorectal tumor grade assessment and response assessment. However, the current literature does not show whether US elastography can differentiate between tumor and fibrosis. Furthermore, it has not been used yet to estimate the resection margin. Therefore, an ex vivo study was performed to investigate the added value of US B-mode and shear wave elastography (SWE) in the classification of colorectal tumor tissue versus the healthy colorectal wall, fat, and fibrosis and in the estimation of the resection margin. Tissue type classification: An ex vivo study was performed to investigate the ability of US B-mode and SWE to distinguish colorectal tumor tissue from healthy colorectal tissue, fat, and fibrosis. Ultrasound B-mode images and SWE images were retrieved simultaneously from freshly excised CRC tissue with maximal three measurements per specimen. Superpixel segmentation was used to divide the US B-mode and SWE images into groups of pixels that share common characteristics. Subsequently, features of US B-mode and SWE images were extracted per superpixel. The superpixels were labeled with a tissue type using histopathology. The US B-mode and SWE features were used as input for machine learning-based classification algorithms. A classification algorithm was developed using nine selected features (five B-mode, and four SWE features) and a Bagged Trees algorithm. Furthermore, another classification algorithm was developed using a Fine Gaussian SVM (FG-SVM) algorithm. The Bagged Trees algorithm resulted in a Matthews correlation coefficient (MCC) of 0.40 for the test set, an accuracy of 0.86, an area under the curve (AUC) value of 0.83, sensitivity 0.87, and specificity 0.74. The FG-SVM algorithm resulted in an MCC of 0.40 for the test set, an accuracies of 0.89, an AUC value of 0.81, sensitivity of 0.91, and specificity of 0.63. The results demonstrate the capability of the proposed technique for the illustration of tumor tissue location during CRC surgery. In this, the B-mode and SWE images are automatically divided into superpixels which result in the detection of the tumor areas in the complete image. The tissue type classification using both imaging modalities is capable of distinguishing colorectal tumor tissue from neoadjuvant effects (fibrosis) and thereby avoiding positive CRM’s and preserving healthy tissue. Resection margin estimation: The recognition of tumor tissue is the first step, but the estimation of the resection margin is more clinically relevant. Therefore, a second ex vivo study was performed to investigate whether a combination of DRS, US B-mode, and SWE can be used to estimate the resection margin in ex vivo specimens. The study population was a sub-study population of the previous study and consisted of 14 specimens with tumor tissue within 1 cm from the resection plane. Regression analysis was performed using DRS, B-mode, and SWE features to estimate the resection margin. The preliminary results of this study showed that features of the US B-mode and SWE images result in a significantly lower mean absolute error (1.88 mm) of the resection margin than DRS (2.10 mm ) or DRS in combination with B-mode and SWE features (1.96 mm). However, the results of this study are preliminary since limited data was available with successful DRS, B-mode, and SWE measurements of specimens with tumor tissue within 1 cm of the resection plane. 6 Discussion and conclusion: In conclusion, the results of the ex vivo studies in this thesis gave new insight into how US B-mode and SWE can be used to distinguish tumor tissue from fibrosis, the healthy colorectal wall, and fat. Furthermore, it demonstrated how the US techniques can be combined with DRS in the estimation of the resection margin of colorectal tumors. However, optimization of the classification and regression algorithms and future research is needed to investigate whether these techniques can lead to less positive resection margins of rectal tumors while preserving more healthy tissues. |
| Item Type: | Essay (Master) |
| Faculty: | TNW: Science and Technology |
| Subject: | 44 medicine, 50 technical science in general |
| Programme: | Technical Medicine MSc (60033) |
| Link to this item: | https://purl.utwente.nl/essays/86109 |
| Export this item as: | BibTeX EndNote HTML Citation Reference Manager |
Show download statistics for this publication
Show download statistics for this publicationRepository Staff Only: item control page