Accelerated cardiac T1 mapping in four heartbeats with inline MyoMapNet: a deep learning-based T1 estimation approach

Table 1. Results of hyperparameter optimization

Layers	Number of neurons in each layer	Activation function	Batch size	Learning rate	Mean Error of estimated T₁ (ms)
Layers	Number of neurons in each layer	Activation function	Batch size	Learning rate	All pixels	Myocardium	Blood
3	400, 400, 1	Leaky Relu	64	0.01	145.5	-26.7	17.9
4	400, 200, 100, 1	Leaky Relu	32	0.01	145.8	-22.6	9.1
5	400, 400, 200, 100, 1	Relu	64	0.01	176.5	26.2	192.8
6	400, 400, 200, 200, 100, 1	Leaky Relu	64	0.01	111.8	-9.4	-7.9
7	400, 400, 400, 400, 200, 100, 1	Relu	64	0.001	137.6	18.1	60.2

Adam optimizer yields the best result in all experiments. The selected hyperparameters for MyoMapNet are highlighted as bold

ISSN: 1532-429X