Atri-U: assisted image analysis in routine cardiovascular magnetic resonance volumetry of the left atrium

Table 1 Characteristics and use of data samples in chronological order

	Training and evaluation of Atri-U			Evaluation in clinical setting		Evaluation aspect
Sample	A	B	C	D₁	D₂
Ground truth of segmentation, landmarks and ES	TA and CA (100 each)	TA and CA	TA and CA (25 each)	GS	n.a	Atri-U modules
Training of Atri-U	✓	X	X	X	X
Evaluation of modules	X	✓	✓	X	X
Evaluation of \(LA{V}_{\mathrm{max}}\)	X	✓	✓	✓	X	Atrial volume
Proposal of segmentation, landmarks and ES	n.a	n.a	n.a	GS and Atri-U	Atri-U	Clinical value (time and quality)
Independent check	n.a	n.a	n.a	PH	PH	Clinical value (time and quality)
Count	200	50	50	50	100
Sample type	1/3 random subsample, 2/3 selected subsample		Random sample^#	Consecutive sample
Acquisition interval	2014–06/2018		07/2018–06/2019	After 07/2019
Magnetic field strength (1.5/3 Tesla)	150/100		25/25	112/38
Atrial dilatation	64 (26%)		16 (32%)	30 (20%)
Structural heart disease*	189 (76%)		35 (70%)	103 (69%)

The convolutional neural networks were trained on sample A, validated on sample B and Atri-U was finally tested on sample C. The time saving value was elaborated on the partially overlapping samples D₁ and D₂, processed by a senior radiologist and Atri-U, respectively, and rated by a senior cardiologist. *Details on the subtypes of structural heart disease are listed in Additional file 1: Table S2 and Figure S2. ^#While maintaining the ratio of magnetic field strength at 1:1
ES = end-systole, CA, GS, PH, TA = authors initials, \({LAV}_{max}\) = maximum left atrial volume, n.a. = not applicable, ✓ (tick) means available/completed, X means not available/not performed

ISSN: 1532-429X