基于深度学习的肺癌检测方法研究

Lung Cancer Fetection Based on Feep Learning

肺癌作为全球发病率最高和死亡率最高的恶性肿瘤,提高肺癌患者存活率最有效的方法就是及早发现、及早诊断、及早治疗。通过人为的观察CT扫描图像,极易出现漏判、误判的情形,计算机辅助诊断(CAD)具有高准确率、高效率的特点,这里本文提出了一种基于机器学习的肺癌检测方法,以肺部图像数据库联盟(LIDC)作为计算机视觉分析图像模型,通过比较各类有效的图像特征,以图像的LBP直方图的方法来表示训练集以及待检测图像的特征,提取肺癌ROI区域及相应病变的特征,引入CART分类器作为弱分类器,然后通过AdaBoost算法对肺结节进行分类学习,构建分类可疑肺癌的AdaBoost分类器,再采用迁移学习的方法将构建好的分类器模型迁移到实际临床CT肺部影像来帮助模型进行训练学习。通过实验数据发现,以上方法对肺癌的识别率能够93.2%,无论对医生还是患者来说都具有很大的现实意义。

As the malignant tumor with the highest incidence and mortality in the world, the most effective way to improve the survival rate of lung cancer patients is early detection, early diagnosis, and early treatment. By artificially observing CT scan images, misjudgments and misjudgments are prone to occur. Computer-aided diagnosis(CAD) has the characteristics of high accuracy and high efficiency. Here, this paper presents a method for lung cancer detection based on machine learning. As a computer vision analysis image model, the Lung Image Database Consortium(LIDC), by comparing various types of effective image features, uses the LBP histogram of the image to represent the training set and the features of the image to be detected, and extracts the ROI region of lung cancer and the corresponding lesion Introduce the CART classifier as a weak classifier, then classify and study the lung nodules through the AdaBoost algorithm, build an AdaBoost classifier that classifies suspicious lung cancer, and then use the transfer learning method to migrate the constructed classifier model to the actual clinical CT lung images to help the model train and learn. Through experimental data, it is found that the above method can recognize lung cancer at 93.2%, which has great practical significance for both doctors and patients.