SeptoSympto: a precise image analysis of Septoria tritici blotch disease symptoms using deep learning methods on scanned images

Table 3 Summary of correlation results for necrosis and pycnidia detection between expert evaluations and SeptoSympto outputs across multiple assessed datasets

A Necrosis
Dataset	Difference with the dataset 1 used for training	SeptoSympto—Expert 1	SeptoSympto—Expert 2	Expert 1 – Expert 2
2	Same conditions	p < 0.001 ρ = 0.94	p < 0.001 ρ = 0.75	p < 0.001 ρ = 0.74
3	Durum wheat	p < 0.001 ρ = 0.95	p < 0.001 ρ = 0.90	p < 0.001 ρ = 0.95
4	Different growing conditions and scanners	p < 0.001 ρ = 0.96	p < 0.001 ρ = 0.92	p < 0.001 ρ = 0.88
5		p < 0.001 ρ = 0.83
6		p < 0.001 ρ = 0.76

B Pycnidia
Dataset	Difference with the dataset 1 used for training	SeptoSympto—Expert 1	SeptoSympto—Expert 2	Expert 1–Expert 2
2	Same conditions	p < 0.001 ρ = 0.83	p < 0.001 ρ = 0.80	p < 0.001 ρ = 0.95
3	Durum wheat	p < 0.001 ρ = 0.58	p < 0.001 ρ = 0.81	p < 0.001 ρ = 0.80
4	Different growing conditions and scanners	p < 0.001 ρ = 0.69	p < 0.001 ρ = 0.71	p < 0.001 ρ = 0.89
5		p < 0.001 ρ = 0.81
6		p < 0.001 ρ = 0.94

Spearman correlations were used to compare expert assessments and SeptoSympto measurements of necrosis and pycnidia. The Spearman's rank correlation coefficient (ρ) and the p-values (p) are given for each comparison. The datasets were obtained from various conditions as shown in Table 1

ISSN: 1746-4811