Essentials of a Robust Deep Learning System for Diabetic Retinopathy Screening: A Systematic Literature Review

Table 2

Potential sources of algorithm limitations present in the validation datasets of the 13 + 2 included studies. ✓ = yes; ✕ = no; † = combination; ‡ = exact number unspecified; § = unknown.

	Public dataset	Reference standard grader number	DMO detection	>1 camera type	Image resolution (pixels)	Number of fields	Automated image quality assessment	Image curation	Other disease detection
Gulshan et al. [24]	†	7-8	✓	✓	299 × 299	1	✕	✓	✕
Gulshan et al. [32]	✕	3 + ‡	✓	✓	299 × 299	1	✕	✓	✕
Krause et al. [33]	✕	6	✓	§	779 × 779	1	✕	✓	✕
Ruamviboonsuk et al. [34]	✕	2–3	✓	✓	779 × 779	1	✕	✓	✕
Voets et al. [35]	✓	‡	✕	✓	299 × 299	1	✕	§	✕
Ramachandran et al. [36]	†	2+‡	✕	✓	§	1–3	✕	✓	✕
Li et al. [37]	†	1–5	§	✓	299 × 299	§	✕	✓	✕
Gargeya et al. [38]	✓	‡	✕	✓	512 × 512	§	✕	✓	✕
Abràmoff et al. [39]	✓	3	✓	✕	§	1	✕	✕	✕
Verbraak et al. [40]	✕	>2	✕	✕	§	2	†	✓	✕
Abràmoff et al. [41]	✕	‡	✕	✕	§	2	✓	✓	✕
Ting et al. [42]	✕	‡	✓	✓	512 × 512	§	✕	✕	✓
Bellemo et al. [43]	✕	2	✓	✕	512 × 512	1	✕	✓	✕
Rogers et al. [44]	†	>2	✓	✓	§	1	✕	✓	✕
Bhaskaranand et al. [45]	✓	>2	✕	✓	§	3	✕	✓	✕