UNET++ appropriately segments disc tissues with high performance. (A) Mean (averaged across tissues) intersection over union (mIOU) and tissue frequency weighted mean intersection over union. Each dot is one histology section from one rat (n = 11), mean and standard deviation are also plotted. (B) Mean intersection over union for each tissue. Each dot is one histology section from one rat (n = 11), mean and standard deviation are also plotted. (C) Representative H&E images of rat representing healthy (Ci, Di, and Ei), moderate (Cii, Dii, and Eii), and severely degenerated samples (Civ, Div, and Eiv). Region of interest (ROI) shown for moderate (Ciii, Diii, and Eii) and severely degenerated samples (Cv, Dv, and Ev) where modest misclassification of the NP, AF, Endplate, and Granulation tissue occur (Noted with Arrow) (D) Representative H&E images of rat disc tissue with ground truth (GT) annotations with respective ROIs. (E) Representative H&E images of rat disc tissue with model ROIs.
Richard D. Bell & coauthors developed a segmentation model to identify seven distinct disc tissues and to use these segmented areas, along with other derived measures, to estimate pathological changes that align with traditional histological scoring: onlinelibrary.wiley....