草榴社区

Accelerating 3D Medical Image Segmentation with Machine Learning

Kerim Genc

Oct 27, 2020 / 6 min read

The trends are clear regarding 3D medical imaging — the demand for high-quality 3D imaging and its processing is increasing while the number of medical professionals is declining. acknowledges that the U.S. shortage of radiologists and other specialist physicians could climb to nearly 42,000 by 2033. Furthermore, research from the journal shows that the number of knee arthroplasties (TKAs) in the U.S. is projected to grow to 3.48?million by 2030, while total hip arthroplasties (THAs) will grow to 572,000 in the same period.

As 3D imaging plays an increasingly integral role in orthopedic procedures, with more and more personalized surgical plans, guides, and devices coming to the market, there is a greater imbalance between the amount of data being created and the scalability of its processing for surgical planning, device/guide design, and even in silico clinical trials.

This imbalance has created an increased demand for automated workflows to reduce the amount of time devoted to 3D image processing. Clinicians, technicians, designers, and engineers need and want to focus more on their final applications, not tedious image processing and anatomical model generation.

The Simpleware? Product Group at 草榴社区 is addressing this need by providing a robust and powerful software platform for 3D image processing and model generation, as well as leveraging machine learning-based artificial intelligence (AI) technology to help automate certain workflows.

Johann Henckel, M.D., an orthopedic surgeon from the Royal National Orthopaedic Hospital, said the following about the Simpleware? platform as it applies to the hospital’s patient-specific workflows:

"Image segementation of MRI and CT scans presents a signifcant challenge for our surgical and egineering multidiscplinary teams. What is currently a laborious process that occupies significant egineering resources and time can now be completed quickly, accurately and with less variability, promising a scalable solution for generating high-fidelity, patient-specific models, surgical tolls and bespoke implants."

Replacement Knee

What Is 3D Image Segmentation?

3D image segmentation is the process of taking 3D data input acquired from various clinical imaging modalities— computed tomography (CT) or magnetic resonance imaging (MRI) — and labeling them to isolate regions of interest such as bone, muscle, and other organs in the human body. Segmented images form the basis of the image-based 3D model, which is a triangulated surface or volume reconstruction. By mapping the regions of interest directly from 3D image data through segmentation, the computer model precisely resembles the patient’s anatomy, as opposed to an idealized version.

Image data also contains information on the condition of bones/soft tissue (depending on modality), providing valuable information to inform decision making. The patient-specific image-based model can then be used for visualization, measurement, planning, design and/or simulation. 草榴社区’ Simpleware? ScanIP software is an industry leader within this area for 3D medical image processing and model generation.

Watch the video below to learn more about Simpleware? software for life sciences applications.

While Simpleware? software provides a comprehensive solution for medical image segmentation, the process can still be quite manual and time-consuming depending on the quality of the image data and the anatomy being segmented. This challenge is being tackled by offering additional machine learning-based AI modules. As illustrated in the figure below, manual 3D image segmentation takes the bulk of time in image-based workflows, with analysis and reporting following at a close second. Any future virtual analysis, such as computer simulation and 3D printing, requires 3D image segmentation.

3D Image Segmentation Chart | 草榴社区

When done manually, a novice segmentation software engineer can take up to three hours to segment the anatomies of one knee MRI scan. Automated segmentation software delivers, on average, a 20 to 50x faster segmentation for clinical images with no user time involved. This software runs unsupervised in just a few minutes, freeing up engineers to devote more time to high-value tasks. The figure below conveys this comparison showing manual segmentation time (left) to fully automated segmentation (right). A video of this comparison can be found .

This automated segmentation technology is fully scalable and consistent for targeting inter-user consistency and repeatability when working with image data. It also streamlines the workflow process of pre-surgical planning and medical device design and permits more automation for in silico trials, increasing the number of participants that can be analyzed in a shorter time.

Machine Learning-Based Automatic 3D Segmentation Tools

Building on a strong track record of life sciences solutions, 草榴社区 offers three automatic 3D segmentation tools that utilize machine learning: Simpleware? AS Ortho, Simpleware? AS Cardio, and Simpleware? Custom Modeler, to shorten segmentation and landmarking time and produce accurate results significantly faster than manual approaches when using medical image data. The Simpleware? AS Ortho (Auto Segmenter for Orthopedics) module as an addition to Simpleware? ScanIP, and segments the hip from CT scans and knees from MRI. In June, a second module, Simpleware? AS Cardio (Auto Segmenter for the Cardiovascular System) was released. This module is used to segment structures of the heart and vascular systems from CT scans.

If the AS modules are not able to provide a 100% automatic segmentation process due to image noise or a significant pathology in the patient, any necessary cleanup of the segmentation can be completed within the ScanIP software environment. This is especially useful in the case of revision surgeries, as discussed in this recent .

The Simpleware? AS Ortho module addresses the anatomy-specific segmentation of the hips and knees. The hip tool works from CT scans to produce a model of the femurs, pelvis, and sacrum, with landmarks placed on the femurs, coccyx, and pelvis. The knee tool works from MRI scans to produce a model of the femur, tibia, and associated cartilage, patella, and fibula, with landmarks placed on the femur and tibia. The figure below shows a 3D knee model.

The Simpleware? AS Cardio module focuses on heart segmentation by segmenting blood pool cavities from each chamber of the heart and selected muscle tissue by identifying landmarks that include key features of the aorta and, specifically, the left ventricle.

In cases where the user needs a solution not currently available in the AS modules (e.g., the shoulder, brain, or additional anatomies/landmarks of the heart), the Simpleware? Custom Modeler module allows customers to work closely with the Simpleware? Product Group to create a fully automated, purpose-built solution tailored to their current process.

Growing Demand for Machine Learning-Based Segmentation

The demand for 3D image-based model generation that utilizes machine learning-enabled segmentation will only increase with time, especially in markets that include patient-specific workflows for medical devices, surgical guides and planning, and in silico clinical trials.

Although it is likely that some form of manual segmentation will be needed for exceptionally difficult cases, the idea of starting a segmentation from scratch will eventually be a thing of the past. Personalized 3D imaging and processing will move customized care forward to provide the most effective clinical applications.

According to the (AHA), patient-centered 3D imaging is undergoing a rapid transformation toward patient-centered medical imaging and an increased emphasis on value and efficient resource allocation. Projections for the market are overwhelmingly positive; indicates that the market for AI-driven image diagnostics will exceed US$3 billion by 2030.

This technology and its many proven applications within life sciences are becoming more entrenched in the medical field, with the ultimate goal of increasingly better patient care.

Continue Reading