Instance-level semantic segmentation of nuclei based on multimodal structure encoding

BMC Bioinformatics

Table 9 Ablation Study on Node Feature Configurations on TCGA-based Cell Classification Dataset

Node features	\(F_{\text {c}\_\text {avg}}\)	\(F_{\text {immune}}\)	\(F_{\text {stroma}}\)	\(F_{\text {tumor}}\)	Feature dim
Visual only	0.840	0.810	0.830	0.840	256
Visual + spatial	0.845	0.815	0.840	0.845	258
Visual + CLIP Text	0.850	0.818	0.860	0.855	768
Visual + spatial + CLIP Text	0.858	0.822	0.875	0.877	770

Visual features are from multi-scale fusion, Spatial features include normalized (x,y) coordinates, and CLIP Text features are extracted from nuclear morphological descriptions

ISSN: 1471-2105