An example of the heatmaps generated by the heatmap generator.
The goal is to create an object detection, classification, and segmentation model that will be able to detect, classify, and segment breast tumors from X-ray images by using several Machine Learning techniques.
A small sample of images is available for experimentation.
- In the folder HeatmapGenerator/Sample_data/
Images, there are 19 sample images for experiments. - In the folder named HeatmapGenerator/Sample_data/
generated_heatmaps, there are generated and saved heatmaps for 316 images.
In the folder Notebook there are jupyter notebook files containing visualizations and evaluations.
- There is a file named Check heatmaps.ipynb in the HeatmapGenerator folder, with which you can check how the Heatmap generator works.
- Evaluate threshold results with bins.ipynb: In this file where we tuned thresholds using bins
- Evaluate threshold results with mean.ipynb: In this file we evaluated the threshold results with mean accuracy scores.
There are also files in which one can perform Grid Search in different ways (without bins, with bins, etc.).
The first approach was taking the MASK R-CNN Segmentation model and replacing its backbone with the ResNet-22 pre-trained on the InBreast mammographic dataset. The backbone and the heatmap generator that are modified and used can be found at Breast Cancer Classifier GitHub Repository.
The next approach was using the F-RCNN CAD model to get a more robust model. The main hardship was installing Caffe, which was eventually successfully completed, in steps described in this file.
The final approach is the one described in this Repository.
- InBreast ~400 .dcm images, 91 of which have extracted bounding boxes
- InBreast ~90 .png images with bounding boxes and masks
- DDSM .tcia file containing ~3500 .dcm images with bounding boxes and masks for the majority of the images
- SFUniversity ~8400 .ljpeg images
- DBT .dcm images, each consisting of ~60 frames
Processed annotations of the above datasets.
@article{
title={Deep Neural Networks Improve Radiologists' Performance in Breast Cancer Screening},
authors={Nan Wu, Jason Phang, Jungkyu Park, Yiqiu Shen, Zhe Huang, Masha Zorin, Stanis\law Jastrzębski, Thibault Févry, Joe Katsnelson, Eric Kim, Stacey Wolfson, Ujas Parikh, Sushma Gaddam, Leng Leng Young Lin, Kara Ho, Joshua D. Weinstein, Beatriu Reig, Yiming Gao, Hildegard Toth, Kristine Pysarenko, Alana Lewin, Jiyon Lee, Krystal Airola, Eralda Mema, Stephanie Chung, Esther Hwang, Naziya Samreen, S. Gene Kim, Laura Heacock, Linda Moy, Kyunghyun Cho, & Krzysztof J. Geras},
year={2019},
doi={https://doi.org/10.48550/arXiv.1903.08297},
}