. 2021 Apr 19;11:8489. doi: 10.1038/s41598-021-86912-w

Table 3.

Cell detection and classification results.

Backend	Immunopositive			Immunonegative			Lymphocyte			Average
Backend	Prec.	Rec.	F1	Prec.	Rec.	F1	Prec.	Rec.	F1	Prec.	Rec.	F1
Modified DeepLabv3-Mobilenetv2⁴⁵	0.8398	0.8252	0.8324	0.7201	0.7468	0.7332	0.1170	0.3471	0.2344	0.7430	0.7606	0.7508
Modified DeepLabv3-Xeption⁴⁵	0.8366	0.8659	0.8510	0.7168	0.8330	0.7705	0.4227	0.4717	0.4458	0.7467	0.8334	0.7871
Modified FCRN-A⁴⁷	0.8287	0.8556	0.8419	0.7190	0.7887	0.7523	0.3950	0.4101	0.4028	0.7449	0.7994	0.7710
Modified FCRN-B⁴⁷	0.8270	0.8726	0.8492	0.7367	0.7908	0.7628	0.4253	0.7376	0.4314	0.7568	0.8069	0.7810
Modified U-Net³⁵	0.8415	0.8436	0.8426	0.7351	0.7899	0.7615	0.4146	0.4630	0.4375	0.7600	0.7972	0.7783
Proposed method	0.8436	0.8611	0.8523	0.7466	0.8198	0.7815	0.3424	0.4246	0.379	0.7766	0.8219	0.7928

Measurements on the test set in terms of precision (Prec.), recall (Rec.), and F1-score (F1.) reported based on different backends. Best results in each column have been shown in bold font.