Table 2.

Enhanced DenseNet169 model structure for skin cancer image analysis.

Stage	Original layer	Enhanced layer	Modification	Parameters (Enhanced)
Input	224 × 224 × 3	224 × 224 × 3	No change	–
Initial convolution	7 × 7 conv, 64 filters, stride 2	3 × 3 conv, 32 filters, stride 2	Reduced filter size & number	~ 0.9 M
Pooling	3 × 3 max pool, stride 2	3 × 3 max pool, stride 2	No change	–
Dense block 1	6 convolutional layers	6 convolutional layers	Retained	~ 1.6 M
Transition layer 1	1 × 1 conv + 2 × 2 avg pool	1 × 1 conv + 2 × 2 avg pool	Retained	~ 0.2 M
Dense block 2	12 convolutional layers	12 convolutional layers	Retained	~ 3.2 M
Transition layer 2	1 × 1 conv + 2 × 2 avg pool	1 × 1 conv + 2 × 2 avg pool	Retained	~ 0.3 M
Dense block 3	32 convolutional layers	16 convolutional layers	Reduced by 50% (less deep features)	~ 1.9 M
Transition layer 3	1 × 1 conv + 2 × 2 avg pool	1 × 1 conv + 2 × 2 avg pool	Retained	~ 0.2 M
Dense block 4	32 convolutional layers	Removed	Significant reduction in computation	0
Final output	Feature vector from final transition layer	Feature vector from final transition layer	Used as input to external classifier	–
Total parameters	≈ 14.3 million	–	~ 50% Reduction	~ 7.35 M