Computer aided diagnosis (CAD) applied to gray-scale ultrasound can help determine which questionable breast masses need biopsy, according to an infoRad display presented by researchers from the University of Chicago.
Although gray-scale ultrasound is 96% to 100% accurate for simple benign cysts, roughly three-quarters of lesions cannot be categorized so easily.
Features of the system include lesion identification, lesion segmentation, feature extraction, and a classifier that yield a diagnosis of the likelihood of malignancy.
The research team collected ultrasound images from 39 patients. Radiologists confirmed benign images by biopsy or cyst aspiration; biopsy also confirmed malignant lesions. Ultrasound scans underwent CAD analysis for lesion margin, texture within the lesion, lesion shape, and posterior acoustic enhancement, particularly acoustic artifacts. Posterior enhancement is more likely with benign lesions, while malignant tumors may show posterior shadowing.
The CAD system then underwent analysis to assess how well it was able to distinguish benign from malignant lesions.
A related scientific exhibit showed the team's efforts with CAD for mammograms. A digital workstation allowed attendees to view up to 30 mammograms, half of which had malignant tumors. The mammograms appeared in random order.
The mammograms had undergone CAD analysis for automatic extraction of image features that were merged with an artificial neural network.
At the exhibit, attendees could view the image in standard and magnified formats. They then guessed whether the patient had at least one lesion and decided whether it was benign or malignant. The computer then showed its estimate, and the viewers could choose to change their opinion. Once a final choice was made, the computer revealed the results of the biopsy.
When the review was finished, the computer printed out a score showing how well the viewer performed compared to other attendees.
Both the infoRad display and the scientific exhibit noted that some members of the team were shareholders in R2 Technology of Los Altos, CA.
