Computer-extracted imaging features that ID prostate cancer: 5 findings

Twitter icon
Facebook icon
LinkedIn icon
e-mail icon
Google icon
 - cancer

Using computer-extracted features to improve prostate cancer diagnosis at MR imaging has powerful potential, but what features are best to discriminate cancer from benign disease?

That was the question that Geert J. S. Litjens, MSc, of Radboud University Medical Center in the Netherlands, and colleagues sought to answer. Their findings, published online in Radiology, could help characterize the role of focal appearance, high-b-value diffusion-weighted images and other features in the diagnosis of prostate cancer.

Litjens and colleagues began with a retrospective cohort of 70 patients, median age of 62, who were scheduled to undergo radical prostatectomy. All underwent preoperative 3T multiparametric MRI, including T2-weighted, diffusion-weighted and dynamic contrast material-enhanced imaging.

A computer application was used to identify features for noncancerous disease categories and prostate cancer.

“The general concept uses image analysis algorithms to extract subvisual image features that are not readily apparent to the human visual system,” wrote Litjens and colleagues. “A good example of this is texture features, which can enhance edges or, conversely, areas of intensity homogeneity.”

The authors reported the following:

  • In distinguishing benign prostatic hyperplasia from cancer, high-b-value diffusion-weighted images were more discriminative than apparent diffusion coefficient.
  • Apparent diffusion coefficient was best used to distinguish prostatic intraepithelial neoplasia from prostate cancer.
  • Looking for roundness and well-defined edges on T2-weighted and contrast-enhanced images helps distinguish inflammatory processes from cancer.
  • Across the imaging techniques, the most useful computer-extracted features depended on cancer grade, with apparent diffusion coefficient the single most important feature do discriminate high-grade cancer from benign lesions.
  • Using only the most clinically relevant features to discriminate benign disease from cancer results in a higher area under the receiver-operating characteristic curve than when all features are used (0.70 vs 0.62)

Noting that the presence of nonmalignant disease such as inflammation or benign prostatic hyperplasia is a common cause of false-positives, Litjens and colleagues stressed that increased understanding of computer-extracted imaging characteristics of benign disease could assist in discriminating these conditions from prostate cancer.

“Our results could be used in two ways,” wrote the authors. “[F]irst, they could form the basis for more granular guidelines for prostate MR imaging interpretation, and second, the results could allow for development of improved computerized decision support systems for diagnosis and characterization of prostate cancer.”