In current medical practice, breast biopsy serves as the gold standard for determining if a breast mass is benign or malignant. Mammography is the main imaging procedure (screening) for detecting suspicious findings with its high sensitivity but its lower specificity (false positives) is well documented. With increased emphasis on early detection of breast cancer, it appears the effort to avoid missing a malignant lesion may have led to a low positive biopsy rate for cancer, between 10-31%. Unfortunately, breast biopsy is neither a benign nor an inexpensive process. Besides affecting patients physically and emotionally, the procedure frequently causes internal scarring, which may obscure the results of future mammograms. With approximately 1,700,000 women undergoing breast biopsy per year in the USA, combined with a cost between $750-5000 per procedure, the cost to the U.S. healthcare system is significant. Ultrasound is widely regarded as the adjunct (secondary) procedure of choice to mammography, especially for distinguishing cystic from solid masses in which accuracy is 96-100%. Results suggest that more accurate application of ultrasound could help reduce the number of biopsies by up to 40% with a cost savings in conservative estimate of well over $1 billion per year in the USA, mainly by reducing the number of False Positives that are biopsied. However, earlier studies in which ultrasound was evaluated largely as a primary screening tool reported a wide variance in Positive Predictive Value (PPV) and an unsettling range of False Negative (FN) rate from 0.3-30%. Despite improving scanner technology with excellent near-field imaging and an increasing number of positive clinical reports that show sonography can distinguish benign from malignant solid nodules with, it is likely that a majority of radiologists in the U.S. recommend that breast US be used only to determine whether a lesion is cystic or solid and/or for needle guidance. Although a well defined rule-based system has been developed (BI- RADS) for reporting, describing and scoring the Level of Suspicion (LOS) for cancer based on the ultrasound appearance of breast masses. Also it is difficult to teach the method with high variability between radiologists. Almen Laboratories with its clinical collaborators has developed a sophisticated computer-aided image analysis system for a variety of medical imaging applications. Almen Labs' Computer-Aided Imaging System (CAIS) provides extensive tools to identify objects of interest or concern such as breast masses in a medical image. It also measures numeric features of the mass, analyzes the important information content and then compares the mass of interest to images of masses with known findings and outcomes. The most similar known masses are retrieved and displayed nearly instantaneously for the radiologist to review. During our Preliminary Study this software system was optimized for the diagnostic breast ultrasound examination to standardize interpretation. Our goals for Accuracy, Sensitivity and Specificity were exceeded and the CAIS also exceeded the performance of four expert radiologists. These results represent the highest performance for any known ultrasound CAD. This Phase I & II Fast Track grant application targets important advancement and more extensive clinical validation of this breast ultrasound tool. The goal is to provide radiologists with a higher degree of confidence to differentiate many benign from more suspicious lesions. We propose to complete evaluation and validation of the Computer-Aided Imaging System in the clinical environment. We hypothesize that correct classification of masses on breast ultrasound (specificity of interpretation) can be significantly improved with no significant reduction in detection of cancer (change in sensitivity) by using the CAIS. Lesions of lower LOS such as complex cystic masses will be ruled out as candidates for biopsy with a higher degree of confidence with use of the computer-aided imaging system. We will acquire approximately 580 new digital ultrasound cases with biopsy confirmation or two-year follow up from the PACS networks at our three medical centers. These IRB approved data will be assembled in an expert research database with physician interpretations and LOS scores for analysis. We will test and enhance the existing CAIS to streamline the graphical user interface and processing flow to meet the needs of the practicing radiologist. We will process and analyze 580 new patient ultrasound images using the created during preliminary studies and "calibrated" and confirmed by leading radiologists template database of 332 biopsy proven cases. In Phase II the impact of Database size and case mix on CAIS accuracy will be statistically evaluated. A group of four expert Radiologists will interpret the 580 validated cases while using the developed CAIS for decision support and their accuracy will be compared to their performance while working alone. Variability and accuracy will be determined using sensitivity and specificity analysis. CAIS reproducibility and sensitivity to imaging system variables also will be examined. The completed CAIS will be deployed to the dedicated Internet portal in order to provide professionals with wide access to the validated system and the database. The developed digital database of lesion templates with known findings also will be available as a "teaching tool." The completed product will help improve accuracy of practitioners, potentially increase their confidence in breast ultrasound particularly in benign findings, reduce variability of interpretations, and increase ultrasound's role in breast cancer detection and management. Significant support from potential commercial partners has already been received. Almen Laboratories and collaborators have developed a sophisticated computer-aided diagnostic system for medical imaging applications that provides extensive tools to identify objects and image features of interest, analyze the information content and then store, retrieve and compare different objects and images of interest based on this information. In preliminary research this software system was tailored to the specific application of standardizing interpretation of diagnostic breast ultrasound. The implemented Computer-Aided Diagnostic tool was designed and tested on an IRB-approved cohort of 332 cases with confirmation of the "truth" via biopsy or benign followup. Sensitivity, Specificity, PPV, NPV and ROC Area (AZ=0.96) exceeded our goals as well as the performance of four radiologists (avg. AZ=0.86) and represents the highest performance for ultrasound CAD in the literature. This Phase I & II Fast Track grant application targets further advancement and performance evaluation of the breast ultrasound computer-aided tool, which may provide radiologists with a higher degree of confidence to differentiate many benign from more suspicious lesions. Results suggest that more accurate application of ultrasound could help reduce the number of biopsies by up to 40% with a cost savings in conservative estimate of well over $1 billion per year in the USA, mainly by reducing the number of False Positives. Although a well-defined rule-based system has been developed (BIRADS) for reporting, describing and scoring the Level of Suspicion (LOS) for cancer based on the ultrasound appearance of breast masses, it is difficult to teach the method and there several studuies show high variability between radiologists. We propose to refine, evaluate and validate our Computer-Aided Imaging System (CAIS) to assist interpretation of breast ultrasound studies in the pre-clinical environment. The System compares a breast mass in question to a database of images with known findings, displays those closest in "Relative Similarity" and computes an estimate of LOS, ultimately with the physician in the loop. Phase I will independently test (using ROC Area) the CAIS on a new set of 580 cases using the Reference Database developed previously. Phase II will evaluate the performance (Sensitivity, Specificity) of the same four Radiologists from our prior study this time using the CAIS as an aid. We hypothesize that the specificity of interpretation of breast ultrasound can be significantly improved with no significant change in sensitivity by following a structured implementation of the ACR BIRADS method for describing and scoring LOS for cancer with our CAIS. Lesions of lower LOS such as complex cystic masses will be ruled out as candidates for biopsy with a higher degree of confidence with use of the computer-aided imaging system. We will acquire approximately 580 new digital ultrasound cases with biopsy confirmation or two-year follow up from the PACS networks at our three medical centers. These IRB approved data will be assembled in an expert research database with physician interpretations and LOS scores for analysis. We will test and enhance the existing CAIS to streamline the graphical user interface and processing flow to meet the needs of the practicing radiologist. A group of four Radiologists will analyze the 580 validated cases using the optimized CAIS for decision support and will be compared to their performance while working alone. Variability and accuracy will be compared using sensitivity and specificity analysis. System reproducibility and sensitivity to imaging system variables also will be examined in detail. The completed CAIS will be deployed to the Internet in order to provide professonals an access to the validated system. The developed digital database of lesion templates with known findings also will be available as a "teaching tool." The completed product will help improve accuracy of practitioners, potentially increase their confidence in breast ultrasound particularly in benign findings, reduce variability of interpretations, and increase ultrasound's role in breast cancer detection and management. Significant support from potential commercial partners has already been received
