人工智慧卷積神經網路應用於骨骼掃描影像判讀之初步成果

近年來人工智慧（artificial intelligence）進步神速，應用這項科技於指紋辨識、人臉辨識、車牌辨識、瞳孔辨識、掌紋辨識、機器視覺、專家系統、自動規劃等等，已經對人類的生活產生了不可思議的影響。應用「卷積神經網路」（convolutional neural network）來判讀核醫科常見的「全身骨骼掃描」影像判讀診斷，國外學者已有幾個初步的研究。本研究取材於國內醫學中心的影像，隨機選出60組影像，其中包括（1）正常影像22組、（2）具有退化性表現的影像25組以及（3）多發性骨骼轉移的影像13組。部分影像用於訓練卷積神經網路，其他的影像用於驗證卷積神經網路判讀診斷的正確性。初步研究結果，判讀影像的正確率為86.7%。相信在不久的將來，人工智慧應用於核醫科常見影像的判讀診斷，將會逐漸開發出驚人的成果。

關鍵字

人工智慧；卷積神經網路；骨骼掃描；影像判讀

並列摘要

In recent years, artificial intelligence gains much progress and its applications are widely applied to finger print recognition, facial recognition, vehicle plate recognition, pupil recognition, palm print recognition, machine vision, expert system, and automatic process planning. There has been several research about using convolution neural network for image interpretation of whole body bone scans. We retrospectively collect 60 whole body bone scintigraphy images in out hospital, including 22 normal images, 25 images with degenerative change, and 13 images with multiple bone metastases. Thirty images were used to train convolution neural network, and other 30 images were used to verify the accuracy of the image classification. The preliminary results showed the accuracy to be 86.7%. We believe that the application of artificial intelligence in image interpretation will gain much progress in the near future.

並列關鍵字

artificial intelligence ； convolution neural network ； bone scintigraphy ； image interpretation

國際替代計量

人工智慧卷積神經網路應用於骨骼掃描影像判讀之初步成果

全文下載

主題瀏覽