TABLE 3.

DenseNet121 architecture with Soft Attention Block for classification of PD and HC cases.

Layers	Output shape	Kernel size and details
Convolution 2D	112×112	7× 7conv, stride 2 (Rui et al., 2019)
Max Pooling 2D	56×56	3×3 max−pool,stride 2
Dense Block (Pahuja et al., 2019)	56×56	$\left[\begin{array}{c}1\times 1\text{conv}\\ 3\times 3\text{conv}\end{array}\right]\times 6$
Transition Layer (Pahuja et al., 2019)	56×56	1× 1conv
	28×28	2×2average pool, stride 2
Dense Block (Prediger et al., 2014)	28×28	$\left[\begin{array}{c}1\times 1\text{conv}\\ 3\times 3\text{conv}\end{array}\right]\times 12$
Transition Layer (Prediger et al., 2014)	28×28	1× 1conv
	14×14	2×2average pool, stride 2
Dense Block (Blesa et al., 2015)	14×14	$\left[\begin{array}{c}1\times 1\text{conv}\\ 3\times 3\text{conv}\end{array}\right]\times 24$
Transition Layer (Blesa et al., 2015)	14×14	1× 1conv
	7×7	2×2average pool,stride 2
Dense Block (Zhou et al., 2009)	7×7	$\left[\begin{array}{c}1\times 1\text{conv}\\ 3\times 3\text{conv}\end{array}\right]\times 16$
Soft Attention Block	7×7	SoftAttention× 1
Classification	1×1	7× 7global average pool
Layer	2	Fully Connected Dense Layer, Softmax