Daniele Regge, M.D.
Computer Aided Diagnosis (CAD) applications in Radiology
High resolution images of the human body are generally obtained by Computed Tomography (CT) or Magnetic Resonance (MR) imaging. CT Colonography is an example of such detailed imaging in which data from a CT scanner are processed by dedicated software to obtain a 3D representation of the colon from inside the intestinal lumen in a similar way to a conventional endoscope view. On the other side, MR imaging has shown promise in localizing cancers, because of its intrinsic high soft-tissue resolution, and combining two or more image modalities can improve the sensitivity of MR tests. However the more variables are introduced the more difficult it is to integrate all available information into one reliable final report, even for an experienced reader. To deal with these complex problems, computer aided diagnosis systems (CAD) have been introduced to help radiologists in diagnosing disease.
The research group has developed a CAD able to produce quantitative maps of the prostate, providing malignancy probability information with useful data that might be difficult to assess visually from MR scans. Regarding the CAD colon, diagnostic performance of CT colonography in individuals at increased risk of colorectal cancer was investigated, and further studies are being conducted using the innovative dual energy technology, to generate so-called virtual non-contrast images. Finally, within the CAD-breast project, we have implemented a fully automatic method for detecting blood vessels in dynamic contrast-enhanced breast MR images. The method could be used to reduce labeling of vascular voxels as parenchymal lesions in the CAD breast, and it could also be used to evaluate the pathological response after neoadjuvant chemotherapy.
Conclusions and perspectives:
Recently, our group has published the results obtained using the CAD system for CT colonography screening. In these studies it has been demonstrated that, at the 6-mm threshold, sensitivity of unassisted reading (79.6%) increased significantly with the use of both second-reader CAD (86.0%) and a double-reading paradigm in which a first-reader CAD is followed by a fast 2-dimensional review (89.2%). Besides, the latter required less reading time than that for second-reader CAD (Δ118 seconds) and was 59 seconds longer than unassisted reading. Conversely, the CAD prostate is still in a preliminary development phase. However, preliminary results obtained on a cohort of 33 patients reported a sensitivity of 94%, with a median number of FP per patient equal to 1, and 2 in the peripheral and central zone, respectively. Finally, regarding the breast project, we obtained promising results in detecting most vessels identified by an expert radiologist. Improving CAD performance will stimulate diffusion of such systems in clinical workflows, helping radiologists in detecting cancer, in particular when diagnosis is time consuming and requires deep expertise. A new branch of research is now focusing an assessing the role of CAD in supporting prostatic MR-guided biopsy, and in estimating tumor aggressiveness by integrating the Gleason Score with imaging, in order to better stratify patients and personalize treatments.