Home          People          Resources          Publications          Projects

Computer Aided Diagnosis | Virtual Surgery | Computer Assisted Surgery | High Performance Computing

Computer Aided Diagnosis
Segmentation of vertebral column from MR scans

Sagittal and axial MR scans of the backbone are analyzed to extract the vertebral column. The method is a combination of active contours, snakes, and region growing.


Segmentation of liver from CT images

The abdominal area contains several organs in close proximity exhibiting similar image characteristics. We use Markov Random Fields to obtain an initial contour of the liver. Gradient vector fields (GVF) and active contours refine this initial estimate and segment the liver.

The project has also lead to the development of a dataset containing abdominal CT scans of 50 patients, with coordinates for the liver boundary. The dataset will be publicly distributed with software to provide similarity metrics, and our segmentation algorithm.



Virtual Surgery

The CADI group is working with the Department of Orthopaedics at UB on simulation-based training modalities for orthopaedic surgery. Orthopaedic surgery deals with complex musculoskeletal structures and mechanical instruments. Algorithms and compute platforms that are perfect to simulate minimally invasive procedures do not scale to orthopaedic surgery. For instance, real-time simulation of the musculo-skeletal structure of the knee will need compute power in excess of 540 TFlops, while the fastest super computer has a sustained throughput of 600 TFlops.

Our research group is working on novel algorithms that can be scaled to High-performance computing (HPC) architectures. We also research new HPC architectures using platforms like the NVIDIA CUDA, and the IBM cell BE. Current thrust areas include:

  • Porting of segmentation and FEM simulation algorithms to HPC platforms
  • Developing HPC architectures using NVIDIA CUDA, IBM Cell, and regular AMD/Intel platforms
  • Performance analysis of algorithms on different HPC platforms
  • Integrating FEM models with haptic devices
  • Tracking medical devices using Infra-red cameras


Computer Assisted Surgery

CADI has developed an image guided neurosurgery toolkit to produce optimum plans resulting in minimally invasive surgeries. The Computer Assisted Surgery (CAS) engine covers several research and engineering solutions.

Finite Element Modeling (FEM) to predict brain shift:

FEM is used to predict intraoperative brain shift during neurosurgery; the system uses a three-dimensional (3D) patient-specific finite element (FE) brain model with detailed anatomical structures using quadrilateral and hexahedral elements.

Methods: A template-based algorithm was developed to build a 3D patient-specific FE brain model. The template model is a 50th percentile male FE brain model with gray and white matter, ventricles, pia mater, dura mater, falx cerebri, tentorium cerebelli, brainstem and cerebellum. Two patient specific models were constructed to demonstrate the robustness of this method. Gravity-induced brain shift after dura opening was simulated based on one clinical case of computer assisted neurosurgery for model validation. The pre-operative MR images were updated by the FE results, and displayed as intraoperative MR images easily recognizable by surgeons.

A set of algorithms for developing a 3D patient-specific FE brain model have been developed. Gravity-induced brain shift can be predicted by this model and displayed as high resolution MR images. Such strategy can be used for not only intraoperative MRI updating, but also pre-surgical planning.



We developed DICOMBox tool based on the DICOM processing algorithm in Eview project, which can view and edit the Dicom images on hand held devices. This work shows the promising future to move computing non-intensive functionalities of the CAS Engine to hand held platform. In terms of the secure access for the CAS Engine, a location based access control model is proposed as a comprehensive solution for CAS Engine to meet the HIPAA standard.


The CAS Database set up in a secure Client/Server architecture allows users to upload case information, image data, planning and annotation information. The system supports several types of navigational queries that assist a surgeon in decision making.

Identify/design and develop advanced (3D) interfaces for navigational queries

The surgical interface will also allow users to navigate possible surgical trajectory even before entering the OR. This is accomplished using a new indexing structure developed by over the course of the CAS program. Called the target tree, this index is a variable height tree that recursively decomposes the search space around a single target point. The index allows for insertion and deletion operations to be intermixed with searches. The target point of the index is the end goal of a surgical procedure, usually a tumor that must be removed.

Augmented Reality

We have successfully developed and implemented a prototype for Augmented Reality (AR) system to visualize invisible critical structures of brain in the real view of patient phantom.

Landmark-based Patient & Atlas Co-Registration

The transfer of anatomical knowledge from 3D atlases to patient images via image-atlas co-registration is a very helpful tool in applications such as diagnosis, therapy planning, and simulation. However, there are anatomical differences among individual patients that make registration difficult; accurate voxel-wise fusion of different individuals is an open problem. For planning and simulation applications accuracy is essential, because any geometrical deviation may be harmful to a patient.

Landmarks-based registration is one of the most popular algorithms in atlas-based application. We have implemented landmarks based registration as our first atlas registration algorithm. Here, AC, PC, L, and R were chosen as our control points.


CADI group has worked on mainly five rigid registration algorithms and a deformable registration technique. Following are the registration techniques:

  • Multi-Resolution Mutual Information
  • Mutual Information
  • Landmark based rigid registration
  • Landmark with Mutual Information.

The concentration has been to achieve best results with minimal time take for registration or fusion of mutli-modality data.

Figure shows the registration result using algorithm “Landmark + Mutual information” and a simple image fusion.


High Performance Computing

Link to Accelerated Computing Lab

The computational requirement of simulation is a combination of the data processed in a model and the time in which processing has to be performed. Brain shift prediction, which involves processing 43,540 elements, takes approximately 20 minutes on a system of six machines. A similar model that simulates the physics of human tissue requires 7.5 minutes on a machine capable of performing at 0.6 GFlops. The simulation has to be performed instantaneously in case of a surgical training system, meaning that same amount of processing (0.6 GFlops in 7.5 minutes) has to be performed in 0.5 micro seconds, and the system has to achieve computing in excess of 6540 TFlops. With the current best supercomputer rated at 290 TFlops, dedicated architectures and novel algorithms will be essential to realize trainable surgical simulation in the near future.

The design of computer architectures is dependent on budgetary, space, and application considerations. General-purpose computing platforms provide low to moderate processing power and high flexibility for running scientific applications. Special purpose computing platforms provide high processing power, with very limited flexibility, often focusing on single tasks. Aggregations of general-purpose platforms into clusters and grids provide additional power, with a corresponding increase in programmatic and user complexity. Our objective is to design hardware configurations that will have the benefits of low cost, high computational power, commercial availability, and low electrical power consumption. The hardware configuration will be divided into two groups: (a) floating point accelerators to enable fast physics computation, and (b) graphics processors to enable 3D visualization. Both groups will be associated with development of specialized algorithms and software. The accelerator boards will be in the form of a PCI-X or HTX card with the processors, local memory, and systems interface logic.

GPUs and FPUs are designed to perform specific operations on large amounts of floating point data. In comparison, general purpose processors like the Intel Pentium 4 or AMD Athlon have to perform a wider range of operations. Recently, the performance of GPUs for certain floating point operations has increased exponentially in comparison to the performance of a CPU (Figure 4). While GPUs by themselves do not have the capability of replacing CPUs, it is known that combining GPUs with CPUs on a single machine can provide significant speedup.

(a) GPU performance increase in comparison
to general purpose processors [23]

(b) Proposed computer architecture for TKA training

Figure 4: Integrating GPUs with conventional CPUs to design a high-speed system.

(a) Floating point accelerators (Clearspeed): Floating point processing tasks will be performed
using CPUs from Clearspeed. This CPU provides multiple floating point and accumulator
processors in a single chip, with codes demonstrating 20-25 GFLOP of sustainable performance.
Clearspeed will provide development tools for this processor. Each hardware platform will have
multiple CPUs and memory systems. Each CPU and memory will be independent of the others
on the board. Current plans will use 2 CPUs on each card. The card will interface to a PCI-X or
HTX bus within a host. The host will provide scheduling, IO access, and similar functions. Each
hardware version will use a common PCI board interface.

(b) Graphics processing units (NVidia): We will take advantage of the latest graphics processing units from NVidia, and their Compute Unified Device Architecture (CUDA). Recent NVidia
GPUs are capable of handling both the graphics processing, and data-intensive processing like
physics simulations. For example, the GeForce 8800 GPU has 128 SPs (stream processors), each
of which can be shifted to various operations (vertex, pixel, geometry, or physics) depending on
application demand. The CUDA computing platform allows developers to design code that can
optimally shift the GPU to different operations. The Scalable Link Interface (SLI) is an NVidia
solution that allows users to link multiple identical video cards, and combine the performance of
the GPUs in these video cards.


| Director | UB Home | CSE Home|About CADI