SBIR Phase I: Semi-Autonomous Adaptive Neural and Genetic Segmentation of Medical Images

这个小型企业创新研究(SBIR)第一阶段项目将实施一个医生辅助的实时自适应系统，用于分割3D医学图像数据中的解剖结构。医学图像分割旨在改变解剖结构的表示形式，使其更易于分析。由于生物系统中这些结构的极端可变性，目前使用的特殊手工方法繁琐、耗时且容易出错。图像分割通常不能以编程方式解决。提出的系统是使用并行图形处理单元(GPU)对单个数据进行基于神经网络(NN)的适配，并结合基于遗传算法(GA)的跨GPU内核的适配。该系统将在几秒钟内从医生使用计算机断层扫描(CT)或磁共振成像(MRI)扫描的感兴趣区域建立起对解剖特征的诊断有用的分割。医学成像的快速增长压倒了现有的诊断专家。一种直观而廉价的系统可以快速准确地提供医学图像的诊断相关分割，具有巨大的商业价值。目前，每次扫描需要大约50分钟的手动准备。估计有20%的疾病的诊断和治疗受益于医学成像。新的扫描技术在分辨率上有所提高，但这些技术并没有使分割更容易或更快。建议的方法将使更多的诊断能够在医生直接控制的质量下进行。

This Small Business Innovative Research (SBIR) Phase I project will implement a physician-assisted, real-time adaptive system for the segmentation of anatomical structures in 3D medical image data. Medical image segmentation seeks to change the representation of an anatomical structure, making it more easily analyzed. Because of the extreme variability of these structures in biological systems, current idiosyncratic manual methods currently in use are tedious, time consuming, and error prone. Image segmentation cannot in general be programmatically solved. The proposed system is a Neural Network (NN) based adaptation of the individual data using parallel Graphics Processing Units (GPUs) and coupled with a Genetic Algorithm (GA) based adaptation across GPU cores. The system will build a diagnostically useful segmentation of the anatomical feature within seconds from an area of interest outlined by a physician using a Computed Tomography (CT) or Magnetic Resonance Imaging (MRI) scan.Fast growth in medical imaging overwhelms available diagnosticians. An intuitive and inexpensive system to quickly and accurately deliver diagnostic relevant segmentation of medical images offers tremendous commercial value. Currently, each scan requires approximately 50 minutes of manual preparation. The diagnosis and treatment of an estimated 20 percent of diseases benefit from medical imaging. Newer scanning technologies have increased in resolution, but such techniques have not made segmenting easier or faster. The proposed method will enable more diagnostics to be done with the quality controlled directly by physicians.