Table 2 Architecture of the proposed CNN model.
Layer type | Configuration | Function |
|---|---|---|
Input | 128 × 128 × 3 (RGB image) | Encoded sample image captured from the flow cell |
Conv2D | 32 filters, 3 × 3 kernel, ReLU activation | Detects initial color features linked to reagent-sample interactions |
MaxPooling2D | 2 × 2 | Reduces dimensionality, retains dominant features |
Conv2D | 64 filters, 3 × 3 kernel, ReLU activation | Captures intermediate patterns in color variations |
MaxPooling2D | 2 × 2 | Prevents overfitting, compresses information |
Conv2D | 128 filters, 3 × 3 kernel, ReLU activation | Learns high-level biomarker-specific features |
MaxPooling2D | 2 × 2 | Further reduces noise and improves generalization |
Conv2D | 128 filters, 3 × 3 kernel, ReLU activation | Reinforces key differentiators for regression |
MaxPooling2D | 2 × 2 | Balances feature complexity and efficiency |
Flatten | — | Converts 2D feature maps into 1D input for dense layer |
Dense | 512 units, ReLU activation | Learns nonlinear relations between features and concentration levels |
Output (Dense) | 1 unit, Linear activation (regression output) | Predicts biomarker concentration from image data |
