Saved in:
Bibliographic Details
Main Authors: Chaudhary, Nisha, Ahmad, Tanveer
Format: Recurso digital
Language:
Published: Zenodo 2026
Online Access:https://doi.org/10.5281/zenodo.19551358
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866902216016658432
author Chaudhary, Nisha
Ahmad, Tanveer
author_facet Chaudhary, Nisha
Ahmad, Tanveer
contents <p>This dataset accompanies the study <em>“An explainable AI approach for comprehensive detection and grading in oral cancer”</em>.</p> <p><strong>Contents of this repository:</strong></p> <p><strong>1. predictions-on-test-data/</strong></p> <ul> <li><code>binary_all_folds_predictions.csv</code>: Model predictions for binary classification (OSCC vs normal) on the test dataset.</li> <li><code>multiclass_all_folds_predictions.csv</code>: Model predictions for OSCC subtype classification (WD, MD, PD) on the test dataset.</li> </ul> <p><strong>2. model-weights/</strong></p> <ul> <li>Trained model weights (.pth files) corresponding to cross-validation folds used in the study. These can be used to reproduce model inference using the provided codebase.</li> </ul> <p><strong>3. sample_WSI/</strong></p> <ul> <li>A small subset of de-identified whole-slide images provided for reference and visualization of histopathological patterns.</li> </ul> <p><strong>4. features/</strong></p> <ul> <li><code>Binary_features_resnet50.hdf5</code>: Patch-level feature embeddings extracted using a ResNet50 backbone for binary classification.</li> <li><code>multiclass_features_resnet50.hdf5</code>: Patch-level feature embeddings used for OSCC subtype classification.</li> </ul> <p>These feature files enable efficient reproduction of the MIL inference pipeline without requiring re-processing of whole-slide images, which are large and subject to sharing restrictions.</p> <p><strong>Notes on data availability:</strong><br>The complete dataset, comprising multi-institutional whole-slide histopathology images, is not publicly available due to ethical and institutional restrictions. These data were collected under approved ethics protocols that permit research use but do not allow unrestricted public redistribution.</p> <p>The implementation code for the OralPatho framework is available at:<br><a class="decorated-link" href="https://github.com/NishaChaudhary23/oralpatho" rel="noopener">https://github.com/NishaChaudhary23/oralpatho</a></p> <p> </p>
format Recurso digital
id zenodo_https___doi_org_10_5281_zenodo_19551358
institution Zenodo
language
publishDate 2026
publisher Zenodo
record_format zenodo
spellingShingle Feature embeddings, model weights and sample WSI for "An explainable AI approach for comprehensive detection, and grading in oral cancer"
Chaudhary, Nisha
Ahmad, Tanveer
<p>This dataset accompanies the study <em>“An explainable AI approach for comprehensive detection and grading in oral cancer”</em>.</p> <p><strong>Contents of this repository:</strong></p> <p><strong>1. predictions-on-test-data/</strong></p> <ul> <li><code>binary_all_folds_predictions.csv</code>: Model predictions for binary classification (OSCC vs normal) on the test dataset.</li> <li><code>multiclass_all_folds_predictions.csv</code>: Model predictions for OSCC subtype classification (WD, MD, PD) on the test dataset.</li> </ul> <p><strong>2. model-weights/</strong></p> <ul> <li>Trained model weights (.pth files) corresponding to cross-validation folds used in the study. These can be used to reproduce model inference using the provided codebase.</li> </ul> <p><strong>3. sample_WSI/</strong></p> <ul> <li>A small subset of de-identified whole-slide images provided for reference and visualization of histopathological patterns.</li> </ul> <p><strong>4. features/</strong></p> <ul> <li><code>Binary_features_resnet50.hdf5</code>: Patch-level feature embeddings extracted using a ResNet50 backbone for binary classification.</li> <li><code>multiclass_features_resnet50.hdf5</code>: Patch-level feature embeddings used for OSCC subtype classification.</li> </ul> <p>These feature files enable efficient reproduction of the MIL inference pipeline without requiring re-processing of whole-slide images, which are large and subject to sharing restrictions.</p> <p><strong>Notes on data availability:</strong><br>The complete dataset, comprising multi-institutional whole-slide histopathology images, is not publicly available due to ethical and institutional restrictions. These data were collected under approved ethics protocols that permit research use but do not allow unrestricted public redistribution.</p> <p>The implementation code for the OralPatho framework is available at:<br><a class="decorated-link" href="https://github.com/NishaChaudhary23/oralpatho" rel="noopener">https://github.com/NishaChaudhary23/oralpatho</a></p> <p> </p>
title Feature embeddings, model weights and sample WSI for "An explainable AI approach for comprehensive detection, and grading in oral cancer"
url https://doi.org/10.5281/zenodo.19551358