Classification
ResNet based CNN training for three histology classes.
Histology Image Pipeline Classification and Denoising
A compact research workflow for turning noisy H&E image patches into cleaner model inputs, then comparing CNN classification outcomes across original and filtered datasets.
Denoising
OpenCV filters, DnCNN wrapper, and Restormer outputs.
Evaluation
Confusion matrices, class recall, and accuracy metrics.
Reproducible
Notebook trail, scripts, requirements, and saved weights.
Noisy input
Real denoising
Classification Workflow
End to end path from image folders to saved model outputs.
-
1
Data
Folder structured histology patches are loaded by class.
-
2
Preprocess
Images are resized, normalized, and optionally filtered.
-
3
Model
A ResNet based classifier is trained with early stopping.
-
4
Evaluate
Accuracy and confusion matrices are recorded in notebooks.
-
5
Predict
Saved checkpoints can be reused for classification runs.
Denoising Comparison
Included sample outputs show how each method changes stain texture, edge detail, and artifact patterns.
Model Comparison
The saved notebook outputs compare original images against Gaussian filtered inputs with a sigma 5 preprocessing step.
Class metric
Audit Notes
Key implementation details found across the cloned branch.
Data and classes
Notebooks use three class groups for filtered experiments: Immune Cells, Invasive Tumor Set, and Non Invasive Tumor.
Preprocessing
Classical scripts apply Gaussian or median filters to class folders and write sibling filtered folders for training.
Models
The repo includes CNN classification checkpoints plus Restormer weights and a DnCNN style wrapper for denoising experiments.
Results
Original CNN accuracy is 92.02 percent. Gaussian filtered CNN accuracy is 92.60 percent in the saved notebook run.
Review flag
The Restormer class wrapper has a path prefix to review before reuse, while the script level Restormer run references the saved weights.