Researchers at the University of Michigan Comprehensive Cancer Center are developing software to help radiologists determine whether lumps or nodules in a patient are cancerous or benign.
Not all masses are cancerous. And to make that determination, when a person undergoes a scan to identify a lump or nodule, a radiologist looks at the texture, the borders and the shape of the mass on the scanned image.
The computer program the researchers have developed reads the same scans the radiologist views, and the combined judgement of the computer and radiologist can help detect more cancers.
“From our experiences in evaluating computer-aided diagnosis (CAD) for breast cancer, using computer aids significantly improves the performance of the radiologist in predicting malignancies of the masses. Radiologists with computers are able to detect more cancers than radiologists by themselves. We expect that CAD for lung cancer can achieve similar results,” says Lubomir Hadjiyski, Ph.D., research assistant professor of Radiology at the U-M Medical School.
In the study, researchers looked at 41 CT scans that showed nodules in the lungs. Current scans and previous scans were fed through a computer program specially designed by the UM researchers to evaluate the size, texture, density and change over time of the nodules. Based on that information, the computer determines how likely the nodule is cancerous.
Previous attempts at computer-aided diagnosis have the computer analyze only the current scan. By allowing the computer to read and compare a series of scans, it gets a complete picture and has the same information the radiologist has.
A CAD system is designed to provide a second opinion to radiologists. The computer analyzes the images with computer-vision techniques specially designed for a given type of cancer or disease. At the same time, the radiologist examines the images and evaluates the likelihood of cancer. The radiologist then compares the two results and makes a final decision.
In many cases, the computer and the radiologist might come to the same conclusion. In other cases, though, the computer may determine a low rate of malignancy for a patient where the radiologist is on the fence. This could tip the scale against performing a biopsy. And if there’s a big difference between the radiologist’s judgement and the computer’s, the patient can be called back for a second look.
“The radiologist is not perfect and the computer is not perfect, but working together they detect more cancers,” says Hadjiyski.
Hadjiyski and his team have developed a similar program to detect breast cancer, and initial testing there is promising. The computer program for both lung and breast cancer needs FDA approval before it can be offered clinically. Hadjiyski stresses that computers will never replace the radiologist entirely but that the technology is meant to complement the radiologist’s judgement.
The one flaw with the computer-aided system is it may return false positive results, identifying masses as cancerous when they are benign. Hadjiyski notes, though, that overall the system detects more cancers. As the researchers fine-tune the technology, they hope to see fewer false positives, and may actually help radiologists identify benign lesions and reduce the number of people undergoing biopsies. Researchers hope next to develop a system that will both detect a lesion and identify it as malignant or benign.