APPLICATION OF LINEAR REGRESSION METHOD FOR ANALYSIS OF CYTOLOGICAL IMAGES QUANTITATIVE CHARACTERISTICS

Abstract

This ar­ticle analyzes the pat­ho­lo­gi­cal con­di­ti­ons of the bre­ast ba­sed on the study of cyto­lo­gi­cal ima­ges. Cyto­lo­gi­cal ima­ges are a se­pa­ra­te class of bi­ome­di­cal ima­ges and are used in the di­ag­no­sis of can­cer. For di­ag­no­se pre­can­ce­ro­us and can­ce­ro­us con­di­ti­ons and tre­at­ment tac­tics, di­ag­nosti­ci­ans use cyto­lo­gi­cal, his­to­lo­gi­cal, and im­mu­no­his­toche­mi­cal ima­ges. For au­to­ma­ting the pro­cess of di­ag­no­sis in on­co­logy, au­to­ma­ted mic­roscopy systems are used. Au­to­ma­ted mic­roscopy systems use com­pu­ter vi­si­on al­go­rithms. Re­cently, mac­hi­ne le­ar­ning al­go­rithms ha­ve be­en used to clas­sify ima­ges. Mic­rosco­pic ima­ge pro­ces­sing is a complex and ti­me-con­su­ming pro­cess, as the ima­ges are cha­rac­te­ri­zed by high no­ise le­vels and the ab­sence of cle­ar con­to­urs of cell nuc­lei. To cal­cu­la­te the qu­an­ti­ta­ti­ve cha­rac­te­ris­tics of cell nuc­lei cyto­lo­gi­cal ima­ges, the met­hod for cal­cu­la­ting the qu­an­ti­ta­ti­ve cha­rac­te­ris­tics of cell nuc­lei ba­sed on ima­ge fil­te­ring al­go­rithms and the­ir au­to­ma­tic seg­menta­ti­on has be­en de­ve­lo­ped. An U-Net con­vo­lu­ti­onal neu­ral net­work archi­tec­tu­re has be­en de­ve­lo­ped for cell nuc­le­us seg­menta­ti­on. In this work, the met­hod of pro­ces­sing cyto­lo­gi­cal ima­ges is de­ve­lo­ped. The met­hod con­sists of six sta­ges. The first step is to lo­ad the ima­ge in­to the com­pu­ters me­mory. In the se­cond sta­ge, the ima­ges are prep­ro­ces­sed. The third sta­ge is the au­to­ma­tic seg­menta­ti­on of ima­ges ba­sed on the con­vo­lu­ti­onal neu­ral net­work of the U-Net type. In the fo­urth sta­ge, the qu­an­ti­ta­ti­ve cha­rac­te­ris­tics of cell nuc­lei are cal­cu­la­ted. In the fifth sta­ge, the qu­an­ti­ta­ti­ve cha­rac­te­ris­tics of the cell nuc­lei are sto­red in a da­ta­ba­se. In the sixth sta­ge, li­ne­ar reg­ressi­on al­go­rithms are used to ob­ta­in qu­an­ti­ta­ti­ve cha­rac­te­ris­tics of cell nuc­lei. Cur­rently, li­ne­ar reg­ressi­on is one of the com­mon appro­ac­hes of mac­hi­ne le­ar­ning to da­ta analysis. In this work, the com­pa­ra­ti­ve analysis of the qu­an­ti­ta­ti­ve cha­rac­te­ris­tics appli­ca­ti­on in cell nuc­lei is car­ri­ed out ba­sed on li­ne­ar reg­ressi­on. The sci­en­ti­fic no­velty of the work is de­ve­lop­ment the met­hod for cal­cu­la­ting the qu­an­ti­ta­ti­ve cha­rac­te­ris­tics of cell nuc­lei, which inclu­des sta­ges of ima­ge fil­te­ring and au­to­ma­tic seg­menta­ti­on ba­sed on the use of a neu­ral net­work such as U-Net. The prac­ti­cal sig­ni­fi­can­ce of the work is the softwa­re imple­men­ta­ti­on of the prep­ro­ces­sing mo­du­les and li­ne­ar reg­ressi­on. In par­ti­cu­lar, in­vesti­ga­ted that the set of pa­ra­me­ters "area, length of the ma­in axis" has 1.4 ti­mes less RMSE er­ror com­pa­red to the set "area, pe­ri­me­ter".