3D Convolutional Neural Network for Lung Nodule Detection

The patent transitions from two-dimensional image block segmentation to three-dimensional volumetric analysis. By processing CT scan data as 3D volumes and using 3D convolutional neural networks, the system can accurately locate nodules anywhere in the lung volume without relying on predefined 2D segmentation blocks, thereby improving accuracy while maintaining computational efficiency.

The patent replaces conventional machine learning algorithms with deep learning-based convolutional neural networks. This substitution enables the system to automatically learn hierarchical features from 3D medical images, achieving superior nodule detection accuracy compared to traditional methods while handling uncertain nodule positions more effectively.

The patent segments the lung volume into multiple 3D slices or volumetric regions that can be processed independently through the convolutional neural network. This segmentation approach maintains processing feasibility by breaking down the complex 3D lung volume into manageable computational units, while the 3D nature of the segments ensures that uncertain nodule positions are captured accurately without causing incorrect recognition.

The patent performs preliminary preprocessing of the CT scan data, including intensity normalization, filtering, and segmentation into lung volumes, before feeding the data into the convolutional neural network. These preliminary actions prepare the data in advance, ensuring that the network receives optimized input that enhances both processing feasibility and recognition reliability, preventing incorrect results due to uncertain nodule positions.