BME2022 Paper Submission & Registration
9th Dutch Bio-Medical Engineering Conference

Powered by
© Fyper VOF.
Conference Websites
10:30   Oncology - I
15 mins
Localization and quantification of droplet vaporization for range verification and dosimetry in proton therapy
Gonzalo Collado-Lara, Sophie Heymans, Marta Rovituso, Yosra Toumia, Hendrik Vos, Jan D'hooge, Nico de Jong, Koen Van Den Abeele, Verya Daeichin
Abstract: Cancer is one of the main causes of death worldwide, and it is estimated that over 50% of the cancer patients will need radiotherapy. Proton therapy is emerging as an advanced radiation therapy, as it provides a better control of the spatial dose deposition than conventional radiotherapy. The stopping position of protons, known as range, can be controlled by tuning their kinetic energy. An increase in the initial kinetic energy of the proton results in an increased range. However, different sources of uncertainty arise in vivo, affecting the proton range and reducing the therapy accuracy. Thus, only with real-time range verification can the full potential of this therapy be exploited. In previous publications, superheated nanodroplets were proposed as a range verification candidate, as their vaporization is triggered by charged particles in a proton beam [DOI: 10.1002/mp.14778]. Here we introduce differential high-frame-rate ultrasound to localize single droplet vaporizations in a proton beam, which are related the proton stopping distribution. Furthermore, we show that the number of vaporizations is related to the droplet volume and concentration. In-vitro experiments were carried out in the research beam line of a clinical proton center (Holland PTC, NL): A polydisperse distribution of perfluorobutane droplets was sorted using differential centrifugation into three subpopulations with different mean-diameter, which were characterized using a Coulter counter (Multisizer 3, Beckman Coulter, USA). Phantoms were prepared with a carbomer matrix that fixed the position of a dispersion of droplets and were irradiated with a pencil proton beam. Simultaneous to the irradiation, ultrasound images were taken at a frame rate of 1000 Hz using a linear probe (L12-5) and a research ultrasound system (Vantage 256, Verasonics, USA). Although the vaporized droplets appeared as clustered scatterers in the B-mode images, event sparsity was achieved subtracting subsequential frames. Vaporization events above the noise level of the system were detected and their position was extracted as the weighted centroid. The proton beam range was independently characterized using an ionization chamber (QubeNext, DE.TEC.TOR, IT) and compared to the vaporization distribution. The spatial distribution of the vaporization events was in excellent agreement with the stopping distribution of protons for both the range and the width. A 0.7 mm mismatch was found in the absolute range, which we attribute to uncertainties within the acoustic setup. After normalizing with the number of protons, the number of vaporizations was proportional to the droplet concentration in the phantom, and quasi-proportional to the droplet volume. These results suggest that high-frame-rate ultrasound is a promising method for in vivo proton range verification, as the proton stopping distribution was directly related to the stochastic distribution of vaporizations. Besides, if the droplet concentration and size is known, the number of vaporizations could be related to the number of protons, which could be further exploited for dosimetry.
15 mins
Deep learning for prostate and zonal segmentation on a multicenter MRI dataset
Hubert Blach, Catarina Dinis Fernandes, Jelle Barentsz, Stijn Heijmink, Hessel Wijkstra, Massimo Mischi, Simona Turco
Abstract: MRI prostate segmentation is an essential step for MRI-TRUS fused guided biopsies and planning of MRI guided prostate cancer (PCa) treatment. Although convolutional neural networks (CNN) have shown great promise for object segmentation tasks in several fields, their application in medical imaging is often hampered by the need for large training datasets. Moreover, an optimal, automated segmentation method should be invariant to the data acquired using different scanners and protocols. In this study, we proposed a CNN for prostate zonal segmentation and evaluated its performance and the robustness towards the acquisition protocol in a small multicenter MRI dataset. The study included MRI examinations (T2w, apparent diffusion coefficient [ADC] and dynamic contrast-enhanced [DCE]) from 75 PCa patients acquired in 3 institutes (Amsterdam UMC [AUMC], Netherlands Cancer Institute [NKI], and Radboudumc [RUMC]), using different scanners (Siemens and Philips), coils and field strengths. Prostate, central gland (CG), and peripheral zone (PZ) were delineated in consensus with a technical expert. A multichannel 2D U-Net was trained using 5-fold cross-validation. The input was constructed by using a combination of T2w, ADC, and DCE images in separate channels. To verify the model’s robustness towards the acquisition protocol, supervised domain adaptation (sDA) was performed. The network was first trained using data from 51 patients from two institutions (19 AUMC and 32 RUMC) and then sDA was performed by re-training with 16 patients, including 8 patients from the third institution (NKI). Both models were evaluated on the same set of 16 unseen NKI patients. Classification performance was assessed using the Dice similarity coefficient (DSC). Significant differences were investigated using Wilcoxon signed-rank test. The network achieved a DSC of 0.90, 0.86 and 0.78 in in the whole organ, the PZ and CG, respectively, before sDA, and a DSC of 0.91, 0.85, and 0.78 in the whole organ, PZ, and CG, respectively, after sDA. No statistically significant differences were found. Despite using a small multicenter dataset, the performance of the two models was comparable to that obtained with larger dataset from a single institute , indicating that training with heterogeneous data facilitates translatability to a new setting while maintaining performance.
15 mins
Self-supervised multi-modality image feature extraction for the progression free survival prediction in oropharyngeal squamous cell carcinoma
Baoqiang Ma, Jiapan Guo, Alessia De Biase, Nikos Sourlos, Wei Tang, Peter van Ooijen, Stefan Both, Nanna Maria Sijtsema
Abstract: Long-term survival of oropharyngeal squamous cell carcinoma patients (OPSCC) is quite poor. Accurate prediction of Progression Free Survival (PFS) before treatment could help in the identification of high-risk patients before treatment which makes it possible to facilitate intensive or less intensive treatments for high- or low-risk patients. In this work, we proposed a deep learning based pipeline for PFS prediction, which is for taking part in the task 2 and task 3 of HECKTOR 2021 challenge (Andrearczyk & al., 2021) (Oreiller & al., 2021). The two tasks both aim at the PFS prediction of OPSCC patients using clinical data and imaging data prior to treatment. In addition to the available CT, PET and clinical data in task 2, GTVs of primary tumors can also be used for PFS prediction in task 3. The proposed pipeline consists of three parts. Firstly, we utilize the pyramid autoencoder for image feature extraction from both CT and PET scans. Secondly, the feed forward feature selection method is used to remove the redundant features from the extracted image features as well as to select relevant clinical features. Finally, we feed all selected features to a DeepSurv (Katzman & al., 2018) model for survival analysis that outputs the risk score on PFS on individual patients. The whole pipeline was trained on 224 OPSCC patients from 5 centres. We have achieved an average C-index of 0.7806 and 0.7967 on the independent validation set for task 2 and task 3 respectively. The C-indices achieved on the test set with 101 patients from 2 centres are 0.6445 and 0.6373, respectively (0.7196 and 0.6978 are the best results from the challenge). It is demonstrated that our proposed approach has the potential for PFS prediction and is feasible for the prediction of other survival endpoints.
15 mins
Evaluation of clinical utility of plasma mutation analyses using droplet digital PCR in lung cancer patients
Esther Visser, Remco de Kock, Sylvia Genet, Ben van de Borne, Maarten Broeren, Federica Eduati, Birgit Deiman, Volkher Scharnhorst
Abstract: Therapies targeting specific genes are beneficial in advanced stage non-small-cell lung cancer (NSCLC) patients harbouring actionable mutations [1]. Therefore, clinical guidelines recommend to test for the presence of mutations in advanced stage non-squamous NSCLC [1]. To identify these mutations, tumour DNA from tissue biopsies, retrieved with invasive procedures, are analysed with Next Generation Sequencing (NGS) in current clinical practice. Alternatively, circulating tumor DNA (ctDNA) from less invasively retrieved plasma could be analysed for these mutations by droplet digital PCR (ddPCR), a faster and cheaper detection method. In order to establish the clinical utility of mutation analysis on plasma using ddPCR, a systemic comparison was performed between the mutations found in plasma by ddPCR to the mutations found in tissue by NGS. In this study, plasma samples of 469 primary lung cancer patients were analysed for mutations in BRAF, EGFR and KRAS using ddPCR multiplex assays [2]. For 159 patients diagnosed with advanced stage non-squamous NSCLC, the plasma ddPCR analysis was compared to tissue NGS analysis available from medical records. Mutations found in plasma by ddPCR had an overlap of 96.6% with mutations found in tissue by NGS. From all mutations found in tissue, 50.9% was detected in plasma with ddPCR. The remaining mutations found in tissue were either unavailable in the ddPCR panel (30.3%) or missed due to lower sensitivity of ddPCR analysis on plasma (18.8%). Additionally, mutations were identified in plasma by ddPCR in 44.1% of patients for whom tissue analyses were recommended by guideline, but with unsuccessful analyses (n = 34) and in 4.3% of the patients outside guideline recommendations (n = 276). Overall, this study shows the potential of ddPCR on plasma to accurately identify a subset of mutations in advanced stage non-squamous NSCLC patients. Plasma ddPCR analysis could potentially be used for screening in which only for ddPCR negatively screened patients, NGS on tissue will be performed. In this way, for less patients NGS analyses would be necessary, resulting in shorter turnaround times, lower costs and reduction of additional tissue biopsies, and for patients unsuitable for invasive procedures additional information on actionable mutations could be obtained. References: [1]: König, D., Prince, S. S., & Rothschild, S. I. (2021). Targeted Therapy in Advanced and Metastatic Non-Small Cell Lung Cancer. An Update on Treatment of the Most Important Actionable Oncogenic Driver Alterations. Cancers, 13(4), 804. [2]: de Kock, R., van den Borne, B., Youssef-El Soud, M., Belderbos, H., Brunsveld, L., Scharnhorst, V., & Deiman, B. (2021). Therapy Monitoring of EGFR-Positive Non–Small-Cell Lung Cancer Patients Using ddPCR Multiplex Assays. The Journal of Molecular Diagnostics, 23(4), 495-505.
15 mins
Towards motion robust MRT during microwave hyperthermia by integrating an 8-channel receiver coil array into the MRcollar
Kemal Sumser, Gennaro G. Bellizzi, Juan A. Hernandez-Tamames, Gerard c. van Rhoon, Margarethus Marius Paulides
Abstract: Introduction: In the head and neck region, precise heating and temperature monitoring is a challenge. The MRcollar, an MR compatible head and neck microwave hyperthermia applicator has been developed for conformal heating and to enable MR thermometry (MRT) during the treatment. MRT has great potential to deliver on temperature monitoring needs but the precision and speed requirements demand high signal-to-noise-ratio (SNR) measurements. Hereto, we developed an integrated applicator-multichannel coil array concept for high SNR performance and parallel imaging by near-skin receiving the radiofrequency signals of the MR imaging [1]. This concept was used in the MRcollar that has twelve antennas for heating and now incorporates eight MR-receive coils. Herein, we quantify the SNR and MRT precision gain that the integrated receive coils (IRC) offer against receiving the signals by the body coil (BC) of the MR scanner. Methods: The IRC array is a combination of three coil arrays: three coils in each half shell and two coils in the head rest. The cylindrical phantom and the standard MRT sequence provided by the MR vendor, i.e. a fast gradient echo sequence was used for the evaluation: TR:100 ms, TE:19.2 ms, flip angle 30°, resolution 1.25x1.25x3 mm3, matrix resolution 256x256. In total 12 acquisitions were made. SNR was calculated in a region of interest (ROI) using the dual image subtraction method [2]. Temporal MRT precision was calculated for each voxel assuming that the temperature remained constant during the experiments. PRFS thermometry was calculated using the first acquisition as the baseline. Results: SNR was found to improve by around 5-fold, i.e. from 23 (BC) to 120 (IRC). This improvement also led an improved MRT precision, i.e. lowest measured standard deviation reduced from 1.8°C (BC) to 0.55°C (IRC) and the average standard deviation in the ROI reduced from 0.91°C (BC) to 0.37°C (IRC). Conclusion: IRC array increased the SNR by 5 times compared to the BC which can be used to half the acquisition times without a further loss in SNR. The MRT precision on average was improved from 0.91°C to 0.37°C. Hence, integration of the receiver arrays close to the skin into MR-hyperthermia devices provides a crucial step towards reliable temperature monitoring during hyperthermia. [1] Sumser K, Bellizzi GG, Forner R, Drizdal T, Tamames JA, Van Rhoon GC, Paulides MM. Dual-Function MR-guided Hyperthermia: An Innovative Integrated Approach and Experimental Demonstration of Proof of Principle. IEEE Transactions on Biomedical Engineering. 2020 Jul 29. [2] Goerner FL, Clarke GD. Measuring signal‐to‐noise ratio in partially parallel imaging MRI. Medical physics. 2011 Sep;38(9):5049-57.
15 mins
State of the art cervical cancer brachytherapy: time action analysis and patient experience
Sharline van Vliet - Pérez, Rosemarijn van Paassen, Linda Wauben, Robin Straathof, Nick van de Berg, Jenny Dankelman, Ben Heijmen, Inger-Karine Kolkman-Deurloo, Remi Nout
Abstract: Objective Brachytherapy (BT) is an important component of the curative treatment for locally advanced cervical cancer. Patient experience in terms of pain, anxiety, and duration of each BT treatment step is still scarcely reported. The aim of this study is to perform a time-action analysis and determine the patient experience during each step of BT treatment of cervical cancer as benchmark and to understand and prioritise further improvements. Method In total 30 patients treated with 69 HDR BT fractions with an intracavitary/interstitial applicator were included for the time-action analysis of which 13 patients (28 fractions) were also included for the patient experience analysis. The time-action analysis included a standardised form with the reported time needed for each step. The patient experience analysis included an EQ-5D questionnaire with health state index (0= dead, 1= full health) and EQ VAS score (0= worst imaginable health, 1= best imaginable health) at the beginning of the day to establish a base line health status, and a numeric rating scale questionnaire (0= perfect situation, 10= worst possible situation) to assess the pain, anxiety and duration experience during each treatment step. The median and interquartile range for all parameters is reported. Results The total procedure time (hours:minutes) from arrival at BT department till discharge was 8:55 (8:00-9:25), which was used for: applicator implantation, recovery from operation, imaging, treatment delivery, applicator removal, and waiting time between each step. At the beginning of the day, patients had a health state index score of 0.82 (0.67-1.00) and EQ VAS score of 0.80 (0.63-0.88). Patients had the highest pain score between imaging and treatment delivery (3 (1-7)), the highest anxiety score during applicator removal (2.5 (0-8)), and the highest duration score between imaging and treatment delivery (6 (0-7.5)). The large variations in scores points at inter- and intra-patient variations. Conclusion This analysis highlights patient experience during different steps of cervical cancer BT workflow. In the future, the time-action and patient experience analysis can be used to optimise different steps of the BT treatment.

end %-->