This project aims to develop an algorithm to localize and identify the orientation of the ablation needle to predict the ablation zone during minimally-invasive tumor ablation procedures.

• Student: Giang Hoang (giang@jhu.edu) BS, MSE Biomedical Engineering, Johns Hopkins Whiting School of Engineering

• Mentor(s): Sheng Xu (xus2@cc.nih.gov), Michael Kassin(michael.kassin@nih.gov), Bradford J. Wood (bwood@cc.nih.gov) NIH Center for Interventional Oncology, Interventional Radiology Section

Radiofrequency Ablation or Microwave Ablation are minimally invasive surgical procedures that utilize the heat generated from electrical energy to destroy cancer cells [1]. The procedure is commonly used to treat small tumors, especially in areas that are close to important anatomical structures [2]. Tumor ablation involves inserting a needle that allows the passage of high-frequency energy, which heats up a targeted area in the tumor and induces coagulative necrosis. The efficacy of the treatment is affected by the impedance and heat regulation of the tissue and most importantly, the location of the ablation needles with respect to the tumor [3]. Thus, it is highly dependent on image guidance to help physicians accurately insert the needles to the ablation site. However, over-treated and under-treated tumors are very common. A 2009 report on liver tumor ablation with RFA showed 9% of cases have major complications from overtreatment and around 40% of cases had undertreated tumors that may result in recurrence [3]. This urges for current image guidance practice to be enhanced in order to improve clinical outcomes in tumor ablation.

Building upon previous work and taking into account the clinical motivation, this project aims to fully implement a CT image segmentation algorithm to localize the ablation needles intraoperatively. With the information extracted from the location of the needle in the image, an approximate ablation zone can be generated and displayed to serve as guidance for the physician. The algorithm developed will be tested on a medical imaging dataset and evaluated for accuracy and efficiency. If successfully implemented, further evaluation can be performed on-site using phantom, with existing clinical instruments and interfaces.

The specific aims of the project are:

1. Localize and identify the orientation of the ablation needles in CT images

2. Generate and superimpose the predicted ablation zone based on the location and orientation of the needles using ellipsoid model

3. Evaluate the accuracy and efficiency of the implemented algorithms

• Minimum: (Expected by 03/15)
  1. MATLAB code and documentation to preprocess Nifti CT images
  2. MATLAB code and documentation to apply an appropriate threshold for needle segmentation

• Expected: (Expected by 03/31)
  1. MATLAB code and documentation to extract needle tip location and needle orientation
  2. Reports on algorithm performance with graphs and statistical analysis by testing on medical images

• Maximum: (Expected by 05/01)
  1. MATLAB code and documentation to predict the ablation zone
  2. Reports on ablation zone prediction performance by testing on phantom/ex vivo samples

• Needle Segmentation and Identify Orientation With MATLAB: This project will first implement medical imaging analysis algorithms to segment the ablation needles and determine their orientation. A block diagram of the technical approach is presented below.

1. The images are preprocessed for standardization and thresholded to obtain a binary mask of all the needles in the images. Note this may include unwanted structure and artifacts.
2. The masked structures will then go through connected components analysis and filtered based on volume and shape to obtain only the objects that represent the ablation needles.
3. Principal component analysis will be applied to obtain the needle orientation.
4. 3D Hough Transform is used to support multiple crossing needles that were not separated during connected component analysis.
5. Once the line segments representing the needles are obtained, the location of the needle tip and the orientation of the needle is determined with respect to the CT coordinate system.

• Ablation Zone Approximation Using Ellipsoidal Approximation Model

1. The volume of the ablation region can be approximated with an ellipsoid using the technical information noted for needle type, pulse type, and amount of energy.
2. Using the location and orientation of the needle, an ellipsoid is superimposed on the CT image for ablation zone visualization.

Preliminary Integration of Ablation Zone Prediction Into Tumor Ablation Procedure:

We conducted an ex vivo experiment using a bovine liver, where we inserted a metal bead to serve as a landmark for a virtual tumor to be generated with varying sizes, shown in image A below. We then insert the ablation needles and take intraoperative CT scans. We processed the intraoperative CT scan using our algorithm and overlaid the result onto the generated tumor image. Image B shows the result for a single needle, and image C shows the result for multiple needles. The ablation zone is in purple and the tumor is in yellow, with incrementing tumor size from left to right (10, 15, and 20 mm in diameter). In this image, we can see that the needles are inserted too deep as compared to the tumor location, which can inform the physician to move the needle back to obtain a better overlap between the ablation zone and the tumor location. Up to now, the implementation stops at this visualization. Next steps, quantitative information such as the percentage of overlap between the tumor and the ablation zone or the distance and direction the needles need to be moved to obtain optimal overlap can be computed and displayed.

Click here for the project timeline.

1. Milestone name: Research Literature
   • Planned Date: 02/20
   • Expected Date: 02/20
   • Status: Completed
2. Milestone name: Obtain and Process Medical Images
   • Planned Date: 02/28
   • Expected Date: 02/28
   • Status: Obtained Phantom Images
3. Milestone name: Implement Needle Segmentation Algorithm
   • Planned Date: 03/12
   • Expected Date: 03/12
   • Status: Completed
4. Milestone name: Extract Needle Location and Orientation
   • Planned Date: 03/19
   • Expected Date: 03/19
   • Status: Completed
5. Milestone name: Test Segmentation Algorithm
   • Planned Date: 03/31
   • Expected Date: 03/31
   • Status: Completed
6. Milestone name: Analyze Testing Result
   • Planned Date: 03/31
   • Expected Date: 03/31
   • Status: Completed
7. Milestone name: Implement Ellipsoid Model
   • Planned Date: 04/20
   • Expected Date: 04/20
   • Status: Completed
8. Milestone name: Overlay Predicted Ablation Zone
   • Planned Date: 04/28
   • Expected Date: 04/28
   • Status: Completed
9. Milestone name: Test and Evaluate Ablation Zone Prediction
   • Planned Date: 05/01
   • Expected Date: 05/01
   • Status: Completed

• Project Plan
  • Project plan presentation
  • Project plan proposal
• Project Background Reading
  • See Bibliography below for links.
• Project Checkpoint
  • Project checkpoint presentation
• Paper Seminar Presentations
  • Original Paper
  • Paper Seminar Presentation
  • Paper Review
• Project Teaser Presentation
  • Teaser Presentation
• Project Final Presentation
  • Final Presentation
• Project Final Report
  • Final Report
  • Google Drive Links to All Documentation, Data, and MATLAB Code.

[1] (ACR), Radiological. “Radiofrequency Ablation (RFA) | Microwave Ablation (MWA) - Liver Tumors”. Radiologyinfo.Org, 2019, https://www.radiologyinfo.org/en/info.cfm?pg=rfaliver.

[2] “Radiofrequency Ablation For Cancer - Mayo Clinic”. Mayoclinic.Org, 2018, https://www.mayoclinic.org/tests-procedures/radiofrequency-ablation/about/pac-20385270.

[3] Egger, Jan et al. “Interactive Volumetry Of Liver Ablation Zones.” Scientific reports vol. 5 15373. 20 Oct. 2015, doi:10.1038/srep15373

[4] Santos, Ricardo S. et al. “Electromagnetic Navigation To Aid Radiofrequency Ablation And Biopsy Of Lung Tumors”. The Annals Of Thoracic Surgery, vol 89, no. 1, 2010, pp. 265-268. Elsevier BV, doi:10.1016/j.athoracsur.2009.06.006. Accessed 25 Feb 2021.

[5] Amalou, H., Wood, B.J. Electromagnetic tracking navigation to guide radiofrequency ablation of a lung tumor. J Bronchology Interv Pulmonol. 2012;19(4):323-327. doi:10.1097/LBR.0b013e31827157c9

[6] Zhou, H., Qiu, W., Ding, M., and Zhang, S., “Automatic needle segmentation in 3D ultrasound images using 3D improved Hough transform”, in Medical Imaging 2008: Visualization, Image-Guided Procedures, and Modeling, 2008, vol. 6918. doi:10.1117/12.770077.

[7] Alpers, J., Hansen, C., Ringe, K., & Rieder, C. “CT-Based Navigation Guidance for Liver Tumor Ablation”. In Eurographics Workshop on Visual Computing for Biology and Medicine. The Eurographics Association. 2017

[8] Wood B.J., Locklin J.K., Viswanathan A, et al. Technologies for guidance of radiofrequency ablation in the multimodality interventional suite of the future. J Vasc Interv Radiol. 2007;18(1 Pt 1):9-24. doi:10.1016/j.jvir.2006.10.013

[9] Zhang, J., Chauhan, S. Real-time computation of bio-heat transfer in the fast explicit dynamics finite element algorithm (FED-FEM) framework, Numerical Heat Transfer, Part B: Fundamentals. 2019; 75:4, 217-238, DOI: 10.1080/10407790.2019.1627812

[10] Bachiller-Burgos, Pilar et al. “A Spiking Neural Model of HT3D for Corner Detection.” Frontiers in computational neuroscience vol. 12 37. 1 Jun. 2018, doi:10.3389/fncom.2018.00037

Google Drive Containing All Documentation, Data, and MATLAB Code.