University of Twente Student Theses

Login

Prognostication from Longitudinal Multisequence Brain MRI using Artificial Intelligence

Loo, Iris van der (2022) Prognostication from Longitudinal Multisequence Brain MRI using Artificial Intelligence.

[img] PDF
12MB
Abstract:Background: Brain tumors are among the deadliest cancers, are difficult to treat and often cause disease or treatment-related side effects. Therefore, evaluating tumor response to treatment plays an important role in treatment decision-making. Magnetic Resonance Imaging (MRI) is the most sensitive modality for the evaluation of brain tumors. Response Assessment in Neuro- Oncology (RANO) criteria are currently the most used to quantify treatment response. Important limitations of the RANO criteria are user-dependency, limited reproducibility and limited use of all information present in the scan. Prognostic artificial intelligence monitoring shows high potential in overcoming these limitations. Methods: As a first step in the conversion to MRI 571 patients were included to make a prognosis based on the fluid-attenuated inversion recovery sequence. The algorithm was adjusted to multiple sequences (n = 576) and externally validated (n = 119). All patients with a primary malignancy of the brain (n = 54) were used to compare to the RANO criteria and volumetric assessments and externally validated (n = 62). Primary outcome measures were the concordance index and Kaplan Meier survival curves. Cox time-varying regression was used for associative analysis. Results: For the fluid-attenuated inversion recovery sequence the logrank test on all three risk groups in the Kaplan Meier survival curves showed a significant difference (p < 0.005) with a concordance index of 0.61. Multisequence analysis of the complete dataset revealed a concordance index of 0.62 with a significant difference between the medium and high-risk patient groups (p < 0.005). The external validation showed a concordance index of 0.55 with a significant difference between medium and high-risk patient groups (p = 0.01). In both datasets the difference between low and medium-risk patients was insignificant. Comparison with other methods showed volumetric assessment had the best prognostic performance with a concordance index of 0.77. Conclusion: This thesis has provided evidence for an alternative method of longitudinal monitoring of cancer patients from brain MRI. Unlike current response evaluation methods, this method works fully automatically and uses all information in the scan including disease or treatment-related side effects. Further improvements are needed to reach equal performance across datasets.
Item Type:Essay (Master)
Clients:
Netherlands Cancer Institute, Amsterdam, The Netherlands
Deventer Ziekenhuis, Deventer, The Netherlands
Faculty:TNW: Science and Technology
Subject:44 medicine
Programme:Technical Medicine MSc (60033)
Link to this item:https://purl.utwente.nl/essays/93163
Export this item as:BibTeX
EndNote
HTML Citation
Reference Manager

 

Repository Staff Only: item control page