Local and Deep Features Based Convolutional Neural Network Frameworks for Brain MRI Anomaly Detection

Table 1

Architecture of the DSRCN.

Layer (type)	Output shape	Param	Connected to
Input layer	(256, 256, 3)	0
Separable conv2d	(256, 256, 64)	259	Input
Batch normalization	(256, 256, 64)	256	Separable conv2d
Separable conv2d	(256, 256, 64)	4736	Batch normalization
Batch normalization	(256, 256, 64)	256	Separable conv2d
Separable conv2d	(256, 256, 64)	4224	Batch normalization
Batch normalization	(256, 256, 64)	256	Separable conv2d
Conv2d	(256, 256, 64)	256	Input
Batch normalization	(256, 256, 64)	256	Conv2d
Max pooling 2d	(256, 256, 64)	0	Batch normalization
Concatenate_1	(256, 256, 128)	0	Max pooling 2d & batch normalization
Separable conv2d	(256, 256, 128)	16640	Concatenate_1
Batch normalization	(256, 256, 128)	512	Separable conv2d
Separable conv2d	(256, 256, 128)	17664	Batch normalization
Batch normalization	(256, 256, 128)	512	Separable conv2d
Separable conv2d	(256, 256, 128)	16640	Batch normalization
Batch normalization	(256, 256, 128)	512	Separable conv2d
Conv 2d	(256, 256, 128)	16512	Concatenate_1
Batch normalization	(256, 256, 128)	512	Conv 2d
Max pooling 2d	(256, 256, 128)	0	Batch normalization
Concatenate_2	(256, 256, 256)	0	Max pooling 2d & batch normalization
Separable conv2d	(256, 256, 256)	66048	Concatenate_2
Batch normalization	(256, 256, 256)	1024	Separable conv2d
Separable conv2d	(256, 256, 256)	68096	Batch normalization
Batch normalization	(256, 256, 256)	1024	Separable conv2d
Separable conv2d	(256, 256, 256)	66048	Batch normalization
Batch normalization	(256, 256, 256)	1024	Separable conv2d
Conv2d	(256, 256, 256)	65792	Concatenate_2
Batch normalization	(256, 256, 256)	1024	Conv2d
Max pooling 2d	(256, 256, 256)	0	Batch normalization
Concatenate_3	(256, 256, 512)	0	Max pooling 2d & batch normalization
Total params: 350,083	Trainable params: 346,499		Nontrainable params: 3,584