Table 2 The tuned Shallow-CNN Structure for Interpretable Tuberculosis Detection (learning rate=0.001, Optimizer = Adam)
From: Explainable deep-neural-network supported scheme for tuberculosis detection from chest radiographs
Layer Type | Output Shape | Number of Kernel | Kernel Size | Stride | Activation |
|---|---|---|---|---|---|
Input Image | 224×224×3 | - | - | - | - |
Convolution-2D-1 | 224×224×32 | 32 | 5×5 | 1×1 | ReLU |
MaxPooling-1 | 112×112×32 | - | 3×3 | 1×1 | - |
Convolution-2D-2 | 112×112×64 | 64 | 3×3 | 1×1 | ReLU |
MaxPooling-2 | 56×56×64 | - | 3×3 | 3×3 | - |
Convolution-2D-3 | 56×56×96 | 96 | 3×3 | 1×1 | ReLU |
MaxPooling-3 | 28×28×96 | - | 3×3 | 3×3 | - |
Convolution-2D-4 | 28×28×96 | 96 | 3×3 | 1×1 | ReLU |
MaxPooling-4 | 14×14×96 | - | 3×3 | 3×3 | - |
Dense-1 | 1×512 | - | - | - | ReLU |
Dense-2 | 1×2 | - | - | - | SoftMax |