We conducted rigorous
validation of the technology in use. During 7 year period four studies and
thousands of patients' cases were processed, scored, analyzed and compared to
biopsy proven findings ("truth"). In the latest (fourth) study the 331 total IRB
approved cases were retrieved chronologically from PACS for patients examined in
2002-2003 with masses with findings known through either biopsy or two-year
follow up. Age range was 40-71 and the database consisted of: 30% simple
cysts, 18% intra-cystic masses, 30% solid benign, 22% carcinoma.
Summary of the Scoring Procedure is as follows:
-
Pre-process Image
-
Segment Suspicious Object (Mass) and confirm by
radiologist
-
Measure Features
-
Compare to Template Database with Known Findings
-
Retrieve & Display Most Similar Masses with biopsy
proven findings
-
Compute BIRADS score based on confirmed pathology
of the retrieved most similar cases
-
Generate Standard BIRADS inspired report
Four sub-specialty breast
imaging radiologists independently scored the US images and were asked to decide
whether to biopsy the mass or not. We appreciate that the radiologist is faced
with a decision-making task during interpretation of breast ultrasound that is
influenced by many factors. The results of accuracy
assessment of our CAD in comparison with 4 radiologists are in the Table below.
