研究目的
To develop a convolutional deep and wide network (CDWN) for automatic lung segmentation in low-dose chest CT scans, improving upon traditional methods that rely on handcrafted features.
研究成果
The proposed CDWN model effectively segments lung regions from chest CT scans with high accuracy and consistency, outperforming traditional and state-of-the-art methods. Its success is attributed to the deep and wide network architecture that learns essential features directly from the data.
研究不足
The performance of the CDWN model depends on the choice of hyper-parameters and the availability of annotated training data. Future work could focus on developing a generic model that automatically finds suitable hyper-parameters and extends to unsupervised learning for unlabeled datasets.
1：Experimental Design and Method Selection
The proposed CDWN model is designed for end-to-end pixel-wise classification on chest CT scans in a supervised manner. It includes an encoder path for feature extraction and a decoder path for dense prediction with spatial features through learnable deconvolutional layers.
2：Sample Selection and Data Sources
The model is trained and evaluated using low-dose chest CT scan images from the LIDC-IDRI database, which includes 3D CT scans from 1018 patients.
3：List of Experimental Equipment and Materials
The model is implemented with Keras in Python, utilizing convolutional and deconvolutional layers with ReLU activation functions and trained with Adam optimizer.
4：Experimental Procedures and Operational Workflow
The training involves initializing kernels and weights with random values, using a batch size of 3 and a learning rate of 0.0001 for 70 epochs. Performance is evaluated using Dice coefficient and accuracy metrics.
5：Data Analysis Methods
The model's performance is compared with state-of-the-art methods using statistical measures such as Dice coefficient, Jaccard coefficient, sensitivity, specificity, accuracy, precision, recall, and f-measure.
独家科研数据包，助您复现前沿成果，加速创新突破
获取完整内容
