Graduate School of Engineering, Mie University Department of Radiology, Mie University School of Medicine Department of Radiology, Mie University School of Medicine Department of Radiology, Mie University School of Medicine Department of Radiology, Mie University School of Medicine Department of Radiology, Mie University School of Medicine Graduate School of Regional Innovation Studies, Mie University
Graduate School of Regional Innovation Studies, Mie University
Proceedings of the Second International Workshop on Regional Innovation Studies : (IWRIS2010)
computerized detection method / clustered micro-calcifications / mammograms / remote image diagnosis / breast cancer screening
A remote image diagnosis via telemedicine net-work is pervasive in Japan because the number of expe-rienced radiologists for reading mammograms in breast can-cer screening is insufficient. However, radiologists in a medical institution where mammograms are sent from the other medical institutions have to read larger number of mammograms, and are overtasked. This would decrease radiologists’ performance. On the other hand, it is known that radiologists’ detection performance will be improved by taking into account the detected lesions automatically by a computerized method. In order to assist radiologists in detec-tion of clustered microcalcifications, therefore, we attempt to introduce our computerized detection method to a remote image diagnosis in Mie Prefecture. For the remote image diagnosis in breast cancer screening, digital mammograms taken at six medical institutions in Mie Prefecture are trans-ferred to a DICOM (Digital Imaging and COmmunication in Medicine) image server in Mie University Hospital via tele-medicine network based on virtual private network. The digital mammograms acquired with five different mammo-graphy equipments which are manufactured by GE (General Electric) Corporation, Fujifilm Corporation, and Konica Minolta are treated in this remote image diagnosis. Our computerized detection method detects individual calcifica-tions on mammograms based on an artificial neural network (ANN) with objective features obtained from a multi-resolution analysis. The proposed method identifies automat-ically which mammography equipment a digital mammogram was acquired with from the DICOM header and uses appro-priate ANN which had been trained for not each mammogra-phy equipment but each vender. We applied the proposed method to all mammograms transferred to the DICOM image server in Mie University in a year. Sensitivity and the num-ber of false positives per image were 94.3% and 0.79 for GE Corporation, 100% and 0.47 for Fujifilm Corporation, and 100% and 0.19 for Konica Minolta, respectively. These re-sults were comparable with the average detection results of the commercial computerized detection systems manufac-tured exclusively for each mammography equipment. The proposed method was shown to have the high detection per-formance for clustered microcalcifications, and to have a possibility of being useful in the remote image diagnosis for breast cancer screening in Mie Prefecture.