Arora Tanvi, Dhir Renu
Department of Computer Science and Engineering, Dr. B.R Ambedkar National Institute of Technology, Jalandhar, Punjab, India.
Med Biol Eng Comput. 2017 May;55(5):733-745. doi: 10.1007/s11517-016-1553-2. Epub 2016 Jul 29.
The genetic defects in the humans are uncovered by studying the chromosomes, as they are the genetic information carriers. They are non-rigid objects and they appear in different orientations when they are imaged. To find out the genetic defects, the chromosomes are pre-processed so that they are not touching, overlapping, and bent, and the noise is also discarded. The presence of bends, overlaps, or touches makes it difficult to uncover the genetic abnormalities. So there is a need for development of an efficient technique to classify the segmented chromosomes into different types and then pre-process them in order to correct their orientation. In this work, a hybrid classification technique based upon correlation-based feature selection and classification via regression approach, which will classify the segmented chromosomes into five categories viz; straight, overlapping, bent, touching, or noise is presented. The performance evaluation has been done using 1592 segmented chromosomes from Advance Digital Imaging Research data set. The over-all accuracy of 94.78 % has been obtained for the five class problem. The performance of the proposed classifier has been compared with Bayes Net, Naïve Bayes, Radial Bias Feed Forward Network, and k-nearest-neighbour classifiers. Based upon this categorization, different pre-processing techniques will be applied to correct the orientation of the chromosomes.
通过研究染色体可发现人类的基因缺陷，因为染色体是遗传信息的载体。它们不是刚性物体，在成像时会呈现出不同的方向。为了找出基因缺陷，对染色体进行预处理，使其不相互接触、重叠和弯曲，同时去除噪声。弯曲、重叠或接触的存在使得难以发现基因异常。因此，需要开发一种有效的技术，将分割后的染色体分类为不同类型，然后进行预处理以校正其方向。在这项工作中，提出了一种基于基于相关性的特征选择和通过回归方法进行分类的混合分类技术，该技术将把分割后的染色体分为五类，即：直形、重叠、弯曲、接触或噪声。使用来自高级数字成像研究数据集的1592条分割染色体进行了性能评估。对于五类问题，总体准确率达到了94.78%。将所提出的分类器的性能与贝叶斯网络、朴素贝叶斯、径向基前馈网络和k近邻分类器进行了比较。基于这种分类，将应用不同的预处理技术来校正染色体的方向。
