Philips Medical Systems has developed software that learns how technologists set up MR scans, then takes over for its human mentor.
Philips Medical Systems has developed software that learns how technologists set up MR scans, then takes over for its human mentor.
The software, called SmartExam, promises improved consistency among specified procedures, while freeing technologists to perform other tasks.
Studies are easier to interpret, because the results from all patients are displayed exactly the same way. Exams conducted on the same patient at different times can be compared more easily, because slices acquired during earlier visits can be matched exactly to those performed later.
"Consistency is important not only in academic settings but in daily practice, because we get used to reading images in a certain way," said Dr. Winfried A. Willinek, senior radiologist in the MRI department at the University of Bonn in Germany and one of the first radiologists in the world to use the software routinely. "Changes in the way the images are displayed make it harder for us."
Willinek and colleagues have been using SmartExam since May 2005, primarily in pediatrics and neurology, particularly in the differential diagnosis of epilepsy. The many and varied angulations needed to evaluate epilepsy patients can be vexing, he said. SmartExam helps.
"Even in these challenging applications, the system works perfectly, and that is why technologists and physicians really like it," he said. "It provides the consistency we need to render the best diagnosis."
The commercial version of SmartExam, introduced by Philips Medical Solutions at the 2005 RSNA meeting, runs on all Philips high-field scanners, including the 3T, 1.5T, and 1T open systems. It is currently compatible only with neuro and orthopedic applications, but the company plans to extend its use for all types of exams.
Philips engineers at the company's research lab in Hamburg, Germany, are now applying the pattern recognition algorithms optimized for these static applications to "time varying anatomies," of which the heart is the most challenging.
"We have started with a version (of SmartExam) that addresses 70% of all exams, but we have set our eyes on the Holy Grail, which is cardiac imaging," said Jacques Coumans, Ph.D., vice president of MR global marketing for Philips Medical Systems.
SmartExam is operational at about 10 institutions, according to Coumans. These have served as beta sites for its development.
"In the beginning, people (at these sites) wondered whether they should give control to the system," he said. "But now they have seen there is complete consistency (from patient to patient)."
SmartExam figures out what the technologist wants to see based on its experience with the operator. For example, brain scans for a particular application might all begin with mid-coronal slices starting after the ventricular system, Willinek said. With experience, the software learns to identify specific anatomic landmarks and set the acquisition accordingly.
The software adjusts for differences due to the way a patient is positioned in the scanner. It also compensates for the anatomic variability that naturally exists from patient to patient.
"Sometimes you have structures missing, for example, in brain aplasia where the corpus callosum is absent, or in brain tumors, where you could have mass effect, causing the midline to shift," Willinek said. "Even in these patients, SmartExam turns out to be very accurate."
SmartExam records changes made by the technologist during the exam-for example, slice thickness, field-of-view, orientation, and angulation-learning preferences along the way.
"It is a teaching process, and after the teaching is done, the automatic planning system of SmartExam does what the technologist used to do," Willinek said.
Radiologists at the University of Bonn completed the training phase for individual applications within the time it took to examine five patients. Preferences, recorded during the learning phase, translate into advantages later on, not only in the reproducibility of exams but in compensating for problems that occur. Willinek recounts one such instance, when SmartExam compensated for patient movement.
"We had to repeat the scout scan, because the system needed to recognize that there was movement of the head," he said. "But with that done, we could identify the same planes and all the pathologies, such as the multiple sclerosis plaques, so SmartExam turned out to be reliable even though there was movement."
SmartExam doesn't necessarily reduce the time of the scan, but it can improve productivity, Willinek said.
"This software changes the work done by the technologist," he said. "Rather than doing the planning, the technologist is able to do other tasks, like postprocessing-and that speeds up the whole process."
Can MRI-Based AI Enhance Pre-Op Prediction of Tumor Deposits in Patients with Rectal Cancer?
October 31st 2024For patients with rectal cancer, an emerging nomogram that combines deep learning and clinical factors had greater than 16 percent and 23 percent increases in accuracy and specificity, respectively, for pre-op prediction of tumor deposits in comparison to clinical factors alone.
Can Diffusion MRI Predict Patient Response to Neoadjuvant Chemotherapy for Breast Cancer?
October 23rd 2024A model emphasizing time-dependent diffusion MRI was 15 percent more effective than apparent diffusion coefficient (ADC) measurements at predicting pathologic complete response to neoadjuvant chemotherapy for women with breast cancer, according to new research.