Pictures of the Future      Fall 2005

It takes powerful software to integrate thousands of slices from high-tech imaging systems into three-dimensional pictures. But the key to supporting physicians in rapidly detecting and characterizing potential pathologies is the use of algorithms based on distilled expert knowledge

Learning-based, intelligent applications are opening a new world of possibilities in healthcare. In twenty years or less unnecessary biopsies could be history, cancer screening will probably be automated, and knowledge- driven (learning-based) software applications will provide clinical decision support with the accuracy of world-class specialists. Treatment planning, implementation and follow-up will be based on the fusion of imaging modalities (see One Plus One is a Lot More) and real-time, knowledge-based characterization and evaluation of the physician’s field of view. Like the power of a gigantic magnifying glass, meticulously sifted data from hundreds of thousands of patient cases will be translated into knowledge, injected into diagnostic systems in the form of algorithms (see Software-Guided Intervention) , and focused on patient exams to produce precision evaluations and ever-improving outcomes.

Tomorrow’s knowledge-based revolution is already being born. Its outlines are visible in northern Italy, where a New Electronic Health Card is creating an information bridge between doctors’ offices, pharmacies and each patient’s medical needs. It’s visible in New York City and Saarbrücken, Germany, where radio-frequency identification wristbands (see Calling all Patients) are seamlessly joining patients with their electronic records. And it’s taking shape in radiology practices around the world, where evolving knowledge-based systems are helping physicians to detect diseases at the earliest possible stage. With this evolving healthcare landscape in mind, Siemens Medical Solutions (SMS) is developing a spectrum of innovative computer-aided detection (CAD) technologies covering multiple imaging modalities and applications.

Catching Lung Nodules Early. Nowhere is the need for knowledge-based healthcare more urgent than in the detection of lung cancer. Not only is this the number one cancer killer worldwide, but a single CT chest scan results in up to 1000 slices—so much information that it places high demands on radiologists’ ability to meet their workloads, which can reach up to 40 scans per day. "The sheer quantity of information being generated by diagnostic systems is making it essential that we implement systems capable of extracting meaningful information from medical data," says Alok Gupta, Ph.D., Vice President for Computer Aided Diagnosis and Therapy at SMS in Malvern, Pennsylvania.

With this in mind, SMS last year launched LungCARE NEV (Nodule Enhanced Viewing), a dedicated application for the localization of small nodules in the lungs. Already available in over 100 clinics worldwide, NEV can detect nodules as small as 3mm in diameter. Like a spell checker in a Word program, NEV can sift through hundreds of images and detect any structures that fit a list of nodule characteristics. As the radiologist analyzes a patient’s lung scan, NEV works as an independent second reader in the background. When the radiologist has completed the initial read, NEV can be activated to highlight potential lesions that may have been missed. "We have found that such techniques help most doctors improve their accuracy by anywhere from 10 to 30 %," says Gupta.

What happens when we think and when we feel? Here, an advanced MR tomography process shows the structure of nerve paths

^{(see also see also Collaboration in Mind)}

Current guidelines established by the U.S. Food and Drug administration (FDA) require NEV to be used as a second reader. But as radiologists gain confidence in the technology, it may eventually be permitted as a concurrent reader or even as a first reader—a step that would dramatically accelerate workflows in radiology departments and allow physicians to focus on the most relevant findings. In fact, the idea of using CAD systems as "first readers" is becoming increasingly attractive in a related field—mammography—where, according to Gupta, "such systems are already detecting above 90 % of anomalies."

When lung nodules are detected, the next step is to help clinicians determine their clinical significance and track findings in followup examinations. Current technology, developed by Siemens Corporate Research (SCR) in Princeton, New Jersey, makes it possible to compare two lung scans to see if a suspected nodule has changed over time. "That’s very difficult and time consuming to do without a powerful IT-based function like CAD," says Gupta. Adds Ingo Schmuecking, M.D., who heads SMS’s marketing activities for the CAD Group, "Clinicians need to be supported in making the best possible decisions for their patients—every time. This means catching cancers as early as possible and avoiding unnecessary biopsies. We believe that CAD has great potential to help by learning to detect changes as clinical experts do, and by learning to combine multiple parameters in order to increase predictive value."

Automated Screening of Colon Polyps? Another CAD area in which SMS and SCR have worked closely is virtual colonography. According to the International Agency for Research on Cancer, colon cancer is the second leading cancer worldwide. Today, the most advanced medical centers perform colon exams using a technique called " virtual flythrough" (Pictures of the Future, Fall 2003, Ultrasound enters the fourth dimension), which was pioneered by Siemens Corporate Research.

But like conventional colonography, flythroughs can miss polyps. With a view to maximizing accuracy, Siemens Medical Solutions recently introduced its syngo Colon-Acography with PEV (Polyp Enhanced Viewing) software, the first FDA-cleared second reader product to support CT colonography. As with NEV, PEV uses algorithms based on expert knowledge, as well as machine learning to detect and analyze features known to be associated with polyps. "The sensitivity of this automated tool is extremely high," say SMS’s Colon CAD Program Manager Luca Bogoni, Ph.D. "In the size range of 6 mm and up—the threshold for danger is around 1 cm—CAD has an accuracy level in the 90s—which is close to that of an expert."

Advanced software is already being used for automated detection of colon polyps in virtual fly‑throughs of the intestine (left image), and in the detection of potential lung nodules (right) in CT data sets

This support will be welcomed by radiologists, especially when virtual colonography as a colon cancer screening test becomes widely accepted and procedure volumes increase. "It’s a question of trust and acceptance," says Gupta. "PEV has the potential to allow radiologists to concentrate their time on pathologies that present tangible health risks."

Toward Data-Based Diagnostics. Whether a radiologist is looking for breast cancer, lung nodules or polyps in the colon, detection is always followed by a critical step: characterization and diagnosis. But determining whether an anomaly may be cancer requires years of practice. Here, even experts rely on second opinions, comparative studies, lab tests and biopsies for answers. Considering the immense cost of such efforts, the U.S.’s National Cancer Institute (NCI) and National Institutes of Health (NIH) recently launched a far-reaching IT initiative for health informatics. An important part of this is a large-scale database to support the development of knowledgedriven diagnostic systems, a project that is strongly supported by SMS. "By pooling the combined expertise of NCI, leading medical institutions and industry, this approach provides uniform and accurate criteria for the development of next-generation diagnostic systems," says Marcos Salganicoff, Ph.D., SMS’ Lung CAD Program Manager. "Active participation in this project will ensure that Siemens is in the forefront of innovation in knowledgedriven medical decision support."

At Siemens, teams specialized in pattern recognition, image processing and data mining are involved in translating early versions of such databases into algorithms that will be the basis of learning systems. "Once the underlying knowledge from thousands of cases has been distilled into algorithms, systems based on the latest machine learning concepts will be able to examine a new case, compare it to a knowledge base, and provide clinical decision support via a probabilitybased determination as to whether a nodule, polyp or microcalcification is cancerous," says SMS’s Schmuecking.

When will the first computer-aided characterization systems become available? No one knows for sure. But with SMS’s purchase of Jerusalem, Israel-based CADVision Medical Technologies in 2004, Siemens took a major step in the direction of providing powerful diagnostic support for physicians in evaluating the 90 million screening mammograms performed each year in the U.S. and Europe. "Mammography was the first application for which CAD was introduced, and it may very well be one of the first where we will see classification technologies," says Gupta.

Another major area of computer-aided detection and classification is cardiac ultrasound. Here, patented algorithms developed at SCR and SMS make it possible to automatically find and track the outlines of a moving heart—something extremely difficult to do with the human eye. The tracking technique is based on the characteristics that top cardiologists look for. "We then teach a program to look for the same things," explains Sriram Krishnan, Ph.D., Cardiac CAD Ultrasound Program Manager at SMS. This helps specialists to quantify the change in volume between systole and diastole (ejection fraction), which is a crucial measure of cardiac health.

Even in CT and MR — diagnostic modalities known for their sharp images—advances in information technology are on the way that will make diagnostic decisions and treatments more precise (Pictures of the Future, Spring 2003, Before Illness Strikes). For instance, separating the contours of a tumor from surrounding tissues is a difficult and time-consuming task. Yet such information is essential when it comes to planning sensitive procedures such as radiation treatment. With this in mind, Prof. Ulrich Lauther, a specialist in mathematical optimization at Siemens Corporate Technology in Munich, has developed a program that could cut the time needed to determine the contours of a tumor from hours to minutes.

Saving Hearts with Image Fusion. Not only is IT transforming diagnostics and medical management, it is also radically changing what happens in the operating room (see Software-Guided Intervention). Take the treatment of cardiac arrhythmias, for instance. Until recently, a cure for this potentially life-threatening condition required open heart surgery. Specialized heart centers have been developing a minimally invasive treatment alternative, that involves advancing an ablation catheter from a small incision at the patient’s groin up into the heart. Correct navigation of the catheter in the heart, however, is a very difficult and lengthy procedure.

But now, thanks to advances in image mapping and registration technologies pioneered by Siemens Corporate Research, the catheter procedure can be conducted much more efficiently and safely. "The key to precise catheter navigation is that we can seamlessly map the catheter into a pre-operative CT image, which shows the patient’s anatomy with a resolution of under 1 mm," explains James Williams, Ph.D., who heads SCR’s Imaging and Visualization Department. Once the correct anatomical location of the arrhythmia has been pinpointed the ablation catheter, which has a heatable tip, burns the errant tissue away. "The procedure is now in clinical use. It is curative, and it has a very high success rate," says Williams.

The importance of information technology in the operating room will continue to grow. Earlier this year, SCR and Johns Hopkins University in Baltimore, Maryland began exploring processing and alignment techniques that could automate the reconstruction of images produced by magnetic resonance (MR) scanning. Given that MR is radiation-free, this could lead to safe, real time, high resolution imaging during the course of an entire operation. Advanced image processing will in turn make it possible to perfectly visualize minuscule catheters and other instruments as they are guided to a target. "In ten years," predicts Williams, "the technology will have reached the point that we will be able to perform what today requires major surgery, such as mitral valve replacement in the heart, using only specialized catheters and real-time magnetic resonance."
                                                                                                                Arthur F. Pease