Technique for Segmentation and Selection of Biomedical Images

Various research fields, such as biology and medicine, have increasingly used techniques for the machine estimation of generated pulses. For example, in biological medicine, classifying information makes it possible to automate the interpretation of incoming data obtained owing to diagnosis, which is necessary when processing large volumes of information arrays. This study aims to develop algorithms that enable the selection of single-type objects in images with subsequent image splitting into a set list of segments with heterogeneous tension, even when the number of reference images is very small. Analysis of existing algorithms allowed us to determine the focus area for developing and improving the efficiency of the algorithms. The existing algorithms show poor performance results in analyzing the dark parts of images, so it makes sense to develop an algorithm for image intensity normalization. The developed algorithm simplifies the procedure of partitioning the training base for the classifier owing to the use of the feature vectors. A random forest algorithm was used for image classification, followed by boundary refinement using a Markov field. The image-splitting algorithm precisely separates parts of the brain structure by applying a Markov field to refine classification results. The proposed classification algorithm showed strong results in comparison with existing algorithms, particularly in the comparison of the Dice criterion. The proposed method shows an average increase of 10% in classification accuracy. One way to improve the presented algorithm is to add texture elements to the feature vector, which allows the identification of distinguishing features of the elements, such as shape and length, which could improve this algorithm for a more accurate classification of substructures. Doi: 10.28991/HEF-2024-05-04-011 Full Text: PDF