modeling_cervical.py

"""
Cervical Cancer Classification Model

Custom CNN model for classifying cervical images into 4 severity classes.
"""

import torch
import torch.nn as nn


class CervicalCancerCNN(nn.Module):
    """
    CNN for cervical cancer classification.

    Classifies cervical images into 4 classes:
    - 0: Normal
    - 1: LSIL (Low-grade Squamous Intraepithelial Lesion)
    - 2: HSIL (High-grade Squamous Intraepithelial Lesion)
    - 3: Cancer

    Args:
        config: Optional configuration dict with keys:
            - conv_layers: List of conv channel sizes (default: [32, 64, 128, 256])
            - fc_layers: List of FC layer sizes (default: [256, 128])
            - num_classes: Number of output classes (default: 4)
            - dropout: Dropout rate (default: 0.5)
    """

    def __init__(self, config=None):
        super().__init__()

        # Default config
        self.config = config or {
            "conv_layers": [32, 64, 128, 256],
            "fc_layers": [256, 128],
            "num_classes": 4,
            "dropout": 0.5,
            "input_channels": 3,
        }

        conv_channels = self.config.get("conv_layers", [32, 64, 128, 256])
        fc_sizes = self.config.get("fc_layers", [256, 128])
        dropout = self.config.get("dropout", 0.5)
        num_classes = self.config.get("num_classes", 4)
        input_channels = self.config.get("input_channels", 3)

        # Build convolutional layers
        layers = []
        in_channels = input_channels

        for out_channels in conv_channels:
            layers.extend([
                nn.Conv2d(in_channels, out_channels, kernel_size=3, padding=1),
                nn.BatchNorm2d(out_channels),
                nn.ReLU(inplace=True),
                nn.MaxPool2d(kernel_size=2, stride=2),
            ])
            in_channels = out_channels

        self.conv_layers = nn.Sequential(*layers)
        self.avgpool = nn.AdaptiveAvgPool2d(1)

        # Build fully connected layers
        fc_blocks = []
        in_features = conv_channels[-1]

        for fc_size in fc_sizes:
            fc_blocks.extend([
                nn.Linear(in_features, fc_size),
                nn.ReLU(inplace=True),
                nn.Dropout(dropout),
            ])
            in_features = fc_size

        self.fc_layers = nn.Sequential(*fc_blocks)
        self.classifier = nn.Linear(in_features, num_classes)

        # Class labels
        self.id2label = {
            0: "Normal",
            1: "LSIL",
            2: "HSIL",
            3: "Cancer"
        }
        self.label2id = {v: k for k, v in self.id2label.items()}

    def forward(self, x):
        """
        Forward pass.

        Args:
            x: Input tensor of shape (batch, 3, height, width)

        Returns:
            Logits tensor of shape (batch, num_classes)
        """
        x = self.conv_layers(x)
        x = self.avgpool(x)
        x = x.view(x.size(0), -1)
        x = self.fc_layers(x)
        x = self.classifier(x)
        return x

    def predict(self, x):
        """
        Predict class labels.

        Args:
            x: Input tensor of shape (batch, 3, height, width)

        Returns:
            Tuple of (predicted_class_ids, probabilities)
        """
        self.eval()
        with torch.no_grad():
            logits = self.forward(x)
            probs = torch.softmax(logits, dim=1)
            preds = torch.argmax(logits, dim=1)
        return preds, probs

    @classmethod
    def from_pretrained(cls, model_path, device="cpu"):
        """
        Load pretrained model.

        Args:
            model_path: Path to model directory or checkpoint file
            device: Device to load model on

        Returns:
            Loaded model
        """
        import os
        from pathlib import Path

        model_path = Path(model_path)

        # Try different file formats
        if model_path.is_dir():
            if (model_path / "model.safetensors").exists():
                weights_path = model_path / "model.safetensors"
                use_safetensors = True
            elif (model_path / "pytorch_model.bin").exists():
                weights_path = model_path / "pytorch_model.bin"
                use_safetensors = False
            else:
                raise FileNotFoundError(f"No model weights found in {model_path}")
        else:
            weights_path = model_path
            use_safetensors = str(model_path).endswith(".safetensors")

        # Create model
        model = cls()

        # Load weights
        if use_safetensors:
            from safetensors.torch import load_file
            state_dict = load_file(str(weights_path))
        else:
            state_dict = torch.load(weights_path, map_location=device, weights_only=True)

        model.load_state_dict(state_dict)
        model.to(device)
        model.eval()

        return model