Automated detection of syncytial aggregates in digitized slide: comparison with pathologist's assessment
H. Pais, R. Shah, D. Roberts , C. Salafia , Placenta , DOI: 10.1016/j.placenta.2017.07.342
Objective: To develop an automated image analysis tool for the identification of the villous syncytium, and the quantitation of different histology features in digitized slides.
Materials and Methods: Three high-resolution 2D images at 4x magnification were taken from one slide from each of the 90 placentas collected as
a part of the Pregnancy, Infection and Nutrition Study, making the total number of images 270. Each image sized between 5-6MB and covered an
area of ~4.3mm2. Definiens software was used to segment the blue knots based on RGB values and then classified them in terms of size (smaller and
greater than 1000 pixels), shape (clumps or long thin) and texture (uniform or intermittent blueness).
Fig. 1, image with few syncytial aggregates, and Fig. 2, image with many syncytial aggregates, Figs.1a, 2a, syncytial aggregates segmented in yellow.
Independent review by two expert pathologists showed a high correlation (r=0.79) with quintile rankings of syncytial aggregates identified by
automated algorithm. Expert pathologist review parsed syncytial aggregates into three categories, syncytial knots with dark nuclei, syncytial sprouts (with variegated nuclei) and synctial bridging. The first two types were easily segmented using image texture/intensity. Syncytial bridges cannot be reliably excluded, but they constitute a small number of identified aggregates in any image.