Development of new diagnostic methods for lung cancer based on computer analysis of ultrasonic B-mode images

本文应用计算机分析经支气管镜检查的肺癌病例的B型图像，对肺内良恶性病变进行鉴别诊断。并将其结果与人体存放的相同用途进行了比较。采用64个中间层单元、50万/100万学习次数和128个中间层单元、50万/100万学习次数的神经网络的诊断准确率为71.0%。百分之六十五点四74.8%。72.0%，重复学习次数之间无显著差异。此外，16年、5年和2年经验的外科医生的诊断准确率分别为69.2%、56.1%和63.6%。人工神经网络的诊断准确性优于外科医生。人工神经网络的计算机图像分析被认为是未来研究的有前途的方法。

Differential diagnosis between benign and malignant lung lesions was made using computer analysis of B-mode images obtained by bronchoscopy for lung cancer cases. The results were compared with human lodgment for the same uses. Diagnostic accuracy of ANN using 64 units in the middle layer and 500,000/1,000,000 learning repetitions and 128 units in the middle layer and 500,000/1,000,000 learning repetitions are 71.0%. 65.4%. 74.8%. 72.0%, no significant difference was found among numbers of learning repetitions. In addition, diagnostic accuracy for the surgeon with 16, 5 and 2 years of experience are 69.2%, 56.1% and 63.6%. Diagnostic accuracy of ANN is better than that of the surgeon. Computer images analysis by ANN was regarded as promising methods for the future study.