Kidney Stone Images Classification

[payalrvs3]

Pull Request for DL-Simplified 💡

Issue Title : Kidney Stone Images Classification

• Info about the related issue (Aim of the project) : Classify CT scan kidney stone images using deep learning - distinguishing small stones from large stones based on bounding box area - and compare four transfer learning models to identify the best approach.
• Name: Payal Sumbhe
• GitHub ID: @payalrvs3
• Email ID: payalrvs310@gmail.com
• Identify yourself: GSSoC '26 Contributor

Closes: #467

Describe the add-ons or changes you've made 📃

Added a complete deep learning project for Kidney Stone Images Classification under Kidney Stone Images Classification/:

• Dataset: YOLO-format CT scan dataset (1,299 images, nc:1, class Tas_Var). Since all images contain stones, classification is based on largest bounding box area with a median split - small (area < 0.0015) vs large (area ≥ 0.0015).
• Models: Xception, InceptionV3, EfficientNetV2S, ConvNeXt-Tiny - all with two-phase transfer learning (frozen base → fine-tune top 30 layers).
• Pipeline: tf.data with AUTOTUNE, caching, augmentation, class weights.
• Evaluation: Accuracy, Precision, Recall, F1-Score, ROC-AUC, Cohen's Kappa.
• Best Model: ConvNeXtTiny - Accuracy: 71.54% | F1: 0.7136 | AUC: 0.7501 | Cohen κ: 0.4302

Type of change ☑️

• Bug fix
• New feature (non-breaking change which adds functionality)
• Code style update
• Breaking change
• This change requires a documentation update

How Has This Been Tested? ⚙️

• Trained and evaluated on Kaggle Notebook using T4 GPU (2×)
• All 4 models trained independently with EarlyStopping and ReduceLROnPlateau
• Evaluated on held-out test set (123 images) using 6 metrics
• Visualisations generated: EDA, bounding boxes, training curves, confusion matrices, ROC curves, comparison chart

Checklist: ☑️

• My code follows the guidelines of this project.
• I have performed a self-review of my own code.
• I have commented my code, particularly wherever it was hard to understand.
• I have made corresponding changes to the documentation.
• My changes generate no new warnings.
• I have added things that prove my fix is effective or that my feature works.
• Any dependent changes have been merged and published in downstream modules.

[github-actions]

Our team will soon review your PR. Thanks @payalrvs3 :)