Please use this identifier to cite or link to this item:
http://hdl.handle.net/11375/27709
Title: | The Application of Artificial Intelligence and Elastography to EBUS-TBNA Imaging Technology for the Prediction of Lymph Node Malignancy |
Authors: | Mistry, Nikkita |
Advisor: | Hanna, Wael |
Department: | Clinical Health Sciences (Health Research Methodology) |
Keywords: | Lung Cancer;Endoscopy;Lymph Nodes;Elastography;Artificial Intelligence |
Publication Date: | 2022 |
Abstract: | Background: Before making any treatment decisions for patients with non-small cell lung cancer (NSCLC), it is crucial to determine whether the cancer has spread to the mediastinal lymph nodes (LNs). The preferred method for mediastinal staging is Endobronchial Ultrasound Transbronchial Needle Aspiration (EBUS-TBNA). However, EBUS-TBNA has been reported to generate inconclusive results as much as 40% of the time. Since this jeopardizes good patient care, there is near-universal consensus on the need to develop and study a novel method for LN staging. Recent research has shown that AI and deep learning are used to accurately interpret images with comparisons to clinicians in radiology, pathology, and cardiology. Additionally, EBUS-Elastography is a novel modality which could be used as an adjunct to EBUS-TBNA for LN staging. This technology uses impedance ultrasonography to measure tissue stiffness. Methods: There are three parts to this thesis. The first part involved the training, validating, and testing NeuralSeg, a deep neural network, to predict LN malignancy based on B-mode EBUS-TBNA images. The second part of this thesis involves EBUS-Elastography, defining the blue colour threshold and the optimal SAR cut-off value based on the blue threshold that most accurately distinguished benign and malignant LN. Finally, this thesis's third part involves validating part II's findings. Results: Part I resulted in an overall accuracy of 80.63% (76.93% to 83.97%), a sensitivity of 43.23% (35.30% to 51.41%), a specificity of 96.91% (94.54% to 98.45%), a positive predictive value of 85.90% (76.81% to 91.80%), a negative predictive value of 79.68% (77.34% to 81.83%), and an AUC of 0.701 (0.646 to 0.755). Part II Level 60 was chosen as the blue threshold with an AUC of 0.89 (95% CI: 0.77-1.00), and the optimal SAR cut off was found to be 0.4959 with a sensitivity of 92.30% (95% CI: 62.10% to 99.60%) and a specificity of 76.50% (95% CI: 49.80% to 92.20%). Using the blue threshold and SAR cut-off, the results of part III resulted in an overall accuracy of 70.59% (95% (CI) 63.50% to 77.01%), the sensitivity of 43.04% (CI: 31.94% to 54.67%), and a specificity of 90.74% (CI: 83.63% to 95.47%). Conclusion: It was observed that both AI and AI-powered EBUS-Elastography achieved high specificities on larger sample sizes, indicative that these methods may be helpful in identifying LN malignancy. However, due to the novelty of these technologies, more extensive multi-centre studies must be conducted before these processes can be standardized. |
URI: | http://hdl.handle.net/11375/27709 |
Appears in Collections: | Open Access Dissertations and Theses |
Files in This Item:
File | Description | Size | Format | |
---|---|---|---|---|
Mistry_Nikkita_FinalSubmission_2022July_MSc.pdf | 9.48 MB | Adobe PDF | View/Open |
Items in MacSphere are protected by copyright, with all rights reserved, unless otherwise indicated.