University of Twente Student Theses
A semi-data-driven approach for automating the assessment of the Spinal Instability Neoplastic Score
Vermeulen, N. (2022) A semi-data-driven approach for automating the assessment of the Spinal Instability Neoplastic Score.
|
PDF
1MB |
| Abstract: | Background context: The Spinal Instability Neoplastic Score (SINS) is a classification system that assesses mechanical stability in metastatic spinal disease. If the final score is ≥7, surgical consultation is recommended to further assess the potential need for stabilization of the spine. However, the radiographic assessment is time-consuming and user dependent. Therefore, there is a need for a solution that assesses the SINS in an automated fashion. Purpose: This study describes the development and evaluation of a semi-data-driven approach for automating the radiographic assessment of the SINS based on PET/CT scans. Methods: 87 PET/CT scans of patients with spinal metastases are included and split into a training(70%), validation(10%), and test set(20%). In this study, a semi-data-driven workflow is developed and assessed consisting of three sequential steps: 1) Three parallel pathways of convolutional neural networks (CNNs) for segmenting/labeling vertebrae, vertebral bodies and spinal metastases respectively from PET/CT scans. 2) Automated extraction of radiomic features. 3) A machine learning model for SINS prediction utilizing the features. Outcome measures: The outcomes of the final model are assessed both as continuous(0-18) by calculating the R2, and binary classes (do/ do not refer) by calculating the sensitivity and specificity. Results: For the final prediction of the SINS, linear regression models showed the best performance (R2 =0.56). As a binary referral tool (do/ do not refer patient), a threshold of 4 (without the pain component) resulted in the highest sensitivity of 0.93 with a corresponding specificity of 0.76. Conclusions: To our knowledge this semi-data-driven approach for SINS classification is the first that shows promising results when it is utilized as a referral tool. However, further optimization and external validation of the model is needed before a reliable conclusion can be drawn. |
| Item Type: | Essay (Master) |
| Clients: | UMC Utrecht, Utrecht, Nederland |
| Faculty: | TNW: Science and Technology |
| Subject: | 44 medicine |
| Programme: | Technical Medicine MSc (60033) |
| Link to this item: | https://purl.utwente.nl/essays/93013 |
| 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