Advanced

AJNR - American Journal of Neuroradiology

Published ahead of print on April 3, 2008
doi: 10.3174/ajnr.A1037

This Article Abstract Full Text Full Text (PDF) Alert me when this article is cited Alert me if a correction is posted Citation Map Services Similar articles in this journal Similar articles in PubMed Alert me to new issues of the journal Download to citation manager Citing Articles Citing Articles via CrossRef Citing Articles via Google Scholar Google Scholar Articles by Yamashita, K. Articles by Honda, H. Search for Related Content PubMed PubMed Citation Articles by Yamashita, K. Articles by Honda, H.

Performance Evaluation of Radiologists with Artificial Neural Network for Differential Diagnosis of Intra-Axial Cerebral Tumors on MR Images

K. Yamashita^a, T. Yoshiura^a, H. Arimura^b, F. Mihara^a, T. Noguchi^a, A. Hiwatashi^a, O. Togao^a, Y. Yamashita^d, T. Shono^c, S. Kumazawa^b, Y. Higashida^b and H. Honda^a

^a Department of Clinical Radiology, Graduate School of Medical Sciences, Kyushu University, Fukuoka, Japan
^b Department of Health Sciences, Graduate School of Medical Sciences, Kyushu University, Fukuoka, Japan
^c Department of Neurosurgery, Graduate School of Medical Sciences, Kyushu University, Fukuoka, Japan
^d Division of Radiology, Department of Medical Technology, Kyushu University Hospital, Fukuoka, Japan

View larger version (34K): [in this window] [in a new window]   Fig 1. Diagram of the basic structure of the ANN. Although only 10 input units and 8 hidden units are shown for illustration, the ANN consists of 15 input units and 9 hidden units.

View larger version (137K): [in this window] [in a new window]   Fig 2. MR images of 2 actual cases. A, Case 1: MR images of a 44-year-old woman with a glioblastoma confirmed on pathologic examination (WHO grade IV). Left image: T2WI shows a heterogeneously hyperintense mass with central necrosis and surrounding signal intensity abnormality likely related to tumor extension and edema. Middle and right images: Precontrast and postcontrast T1WIs show hemorrhagic mass and peripheral enhancement with central necrosis, characteristic of glioblastoma. B, Case 2: MR images of a 62-year-old woman with proved metastatic brain tumor from lung cancer. Left image: T2WI shows a cystic frontoparietal mass with mixed-aged hemorrhage. Middle and right images: Precontrast and postcontrast T1WIs show a thin layer of peripheral enhancement. Surgery disclosed adenocarcinoma.

View larger version (12K): [in this window] [in a new window]   Fig 3. ANN output obtained on the basis of 2 radiologists' ratings of MR features and clinical information for the 2 cases shown in Fig 2. Each graph shows the largest output values among the 4 categories corresponding to the correct diagnoses. A, Case 1: The likelihood of high-grade glioma is very high. ANN led to the correct diagnosis. B, Case 2: The likelihood of metastasis is approximately equivalent to high-grade glioma and malignant lymphoma. ANN might fail to lead to the correct diagnosis.

View larger version (22K): [in this window] [in a new window]   Fig 4. Average AUC values and binormal ROC curves for observers with and without ANN output (averaged plot values for all readers). Those for ANN alone are also indicated. Note that observer performance improves significantly with ANN output.

View larger version (24K): [in this window] [in a new window]   Fig 5. The number of correctly diagnosed cases for which observers' rankings changed because of ANN output. Positive values indicate that ANN was beneficial, whereas negative values indicate that ANN was detrimental. ANN output clearly improved the performance.