Register now After registration you will be able to apply for this opportunity online.
Misestimation of CT-perfusion output in acute stroke due to attenuation curve truncation
In this master's thesis project, we are looking for a candidate to apply machine learning techniques to correct and predict signals of incomplete CT perfusion imaging for ischemic stroke. We hope to use machine learning techniques to de-noise and correct for the truncation in CT perfusion signals. In particular, we aim to infer the true attenuation curve after the truncation time-point.
In this master's thesis project, we are looking for a candidate to apply machine learning techniques to **correct and predict signals of incomplete CT perfusion imaging for ischemic stroke**. When the CT perfusion is correctly performed, the contrast-media attenuation time-curves for artery, vein, and bolus in tissue follow a nice curve of first shooting up, then coming down and shooting up a lower peak again, as the blood follows the circulation from brain to lung and back to brain again. However, during insufficient CT perfusion, the curves are delayed and truncated, resulting in an off-phase pattern shift. This truncation of the attenuation curves leads to an inaccurate estimation of the tissue perfusion parameters, which are used to quantify the ischemic core and the penumbral region (critically hypoperfused brain tissue). As a result, there tends to be misestimation of the perfusion parameters and subsequently of the penumbra region. **This can lead to inaccurate core to penumbra region ratio, and thus may lead to incorrect treatment decisions**.
We hope to use machine learning techniques to de-noise and correct for the truncation in CT perfusion signals. In particular, **we aim to infer the true attenuation curve after the truncation time-point**.
**Related Work**
In 2023, Ezequiel et al. trained a classifier on features derived from CTP scans to detect truncation artifacts (https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0283610) and showed that AIF/VOF based classifiers were more accurate than the scans’ duration for detecting truncation. Also in 2023, Moghari
et al. employed VAE-GAN to generate video frames for the truncated CTP scans (https://iopscience.iop.org/article/10.1088/1361-6560/acdf3a). Previously, our semester project student formulated the problem as a time-series prediction task, and used LSTM based RNN model to predict the truncated signals.
Please check out these references for more context:
1. https://practicalneurology.com/articles/2019-nov-dec/rapid-automated-ct-perfusion-in-clinical-practice
2. https://www.ahajournals.org/doi/full/10.1161/STROKEAHA.121.035049
3. https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0283610
4. https://iopscience.iop.org/article/10.1088/1361-6560/acdf3a
In this master's thesis project, we are looking for a candidate to apply machine learning techniques to **correct and predict signals of incomplete CT perfusion imaging for ischemic stroke**. When the CT perfusion is correctly performed, the contrast-media attenuation time-curves for artery, vein, and bolus in tissue follow a nice curve of first shooting up, then coming down and shooting up a lower peak again, as the blood follows the circulation from brain to lung and back to brain again. However, during insufficient CT perfusion, the curves are delayed and truncated, resulting in an off-phase pattern shift. This truncation of the attenuation curves leads to an inaccurate estimation of the tissue perfusion parameters, which are used to quantify the ischemic core and the penumbral region (critically hypoperfused brain tissue). As a result, there tends to be misestimation of the perfusion parameters and subsequently of the penumbra region. **This can lead to inaccurate core to penumbra region ratio, and thus may lead to incorrect treatment decisions**.
We hope to use machine learning techniques to de-noise and correct for the truncation in CT perfusion signals. In particular, **we aim to infer the true attenuation curve after the truncation time-point**.
**Related Work** In 2023, Ezequiel et al. trained a classifier on features derived from CTP scans to detect truncation artifacts (https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0283610) and showed that AIF/VOF based classifiers were more accurate than the scans’ duration for detecting truncation. Also in 2023, Moghari et al. employed VAE-GAN to generate video frames for the truncated CTP scans (https://iopscience.iop.org/article/10.1088/1361-6560/acdf3a). Previously, our semester project student formulated the problem as a time-series prediction task, and used LSTM based RNN model to predict the truncated signals.
Please check out these references for more context: 1. https://practicalneurology.com/articles/2019-nov-dec/rapid-automated-ct-perfusion-in-clinical-practice 2. https://www.ahajournals.org/doi/full/10.1161/STROKEAHA.121.035049 3. https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0283610 4. https://iopscience.iop.org/article/10.1088/1361-6560/acdf3a
Ideal candidates should have basic knowledge in machine learning, linear algebra, statistics, and calculus and interest in clinical neurosciences. A passion to solve real-world clinical problems such as in this project, ischemic stroke treatment and planning, is preferred! You can learn along the way medical image analysis, CT perfusion signal analysis, and some exciting machine learning for the medical domain.
Ideal candidates should have basic knowledge in machine learning, linear algebra, statistics, and calculus and interest in clinical neurosciences. A passion to solve real-world clinical problems such as in this project, ischemic stroke treatment and planning, is preferred! You can learn along the way medical image analysis, CT perfusion signal analysis, and some exciting machine learning for the medical domain.
Please send your application (a current CV including list of publications and a text summary of the research plan) to
Dr Schubert Tilman (tilman.schubert@usz.ch), Dr
Michels Lars (lars.michels@usz.ch) from USZ and Kaiyuan Yang (kaiyuan.yang@uzh.ch), Dr Ezequiel de la Rosa (ezequiel.delarosa@uzh.ch) from UZH (Prof. Bjoern Menze).
Please send your application (a current CV including list of publications and a text summary of the research plan) to Dr Schubert Tilman (tilman.schubert@usz.ch), Dr Michels Lars (lars.michels@usz.ch) from USZ and Kaiyuan Yang (kaiyuan.yang@uzh.ch), Dr Ezequiel de la Rosa (ezequiel.delarosa@uzh.ch) from UZH (Prof. Bjoern Menze).