Advance Image Analysis & Machine and Deep Learning Final Project

Calcification Detection in Mammography Images

MAIA Master 2022

Team Members

Cortina Uribe Alejandro
Seia Joaquin Oscar
Zalevskyi Vladyslav

Instructions

This repository contains all the code and analysis notebooks for our final project on microcalcifications (MC) detection on mammography images that we develpped for our AIA and ML-DL courses. The project included the development of three different pipelines for MC detection, one based on AIA+ML and two deep learning based. The full report can be checked here.

The main final results analysis notebooks are:

On validaton set:
- Adavanced Image Analysis and Machine Learning [Github]/[Local] [Colab]
- Deep Learning - Detection by classification [Github]/[Local] [Colab]
- Deep Learning - Detection with FasterRCNN [Github]/[Local] [Colab]
- Three pipelines - best models [Github]/[Local] [Colab]
On test set:
- Three pipelines - best models [Github]/[Local] [Colab]

Even if the repository is self-contained and it be fully reproduced following the undegoing instructions, the reader might be interested first in checking and runing examples of the final pipelines we generated. To do so, you should do:

Environmental set up
Download the checkpoints for the models
Download the example image (or provide one of your own)
Run the examples

The mentionned steps are here provided in bash command line format:

Environment set up

Start by creating a new conda environment

conda update -n base -c defaults conda &&
conda create -n calc_det anaconda &&
conda activate calc_det &&
conda install pytorch torchvision torchaudio cudatoolkit=10.2 -c pytorch

Install requirements:

pip install -r requirements.txt

Download checkpoints of pretrained models

Deep learning detection by classification

mkdir deep_learning/classification_models/checkpoints &&
cd deep_learning/classification_models/checkpoints &&
gdown https://drive.google.com/uc?id=16BbvvZcS2Qx421v9QKpKH4JrVKF1Efcf &&
unzip classification_checkpoints.zip &&
rm -rf classification_checkpoints.zip &&
cd ../../../

Deep learning detection with FasterRCNN

mkdir deep_learning/detection_models/checkpoints &&
cd deep_learning/detection_models/checkpoints &&
gdown https://drive.google.com/uc?id=1R8fxd_CdyG5ec1grobRUut8UqCKbVFdp &&
unzip detection_checkpoints.zip &&
rm -rf detection_checkpoints.zip &&
cd ../../../

Machine Learning

mkdir machine_learning/checkpoints &&
cd machine_learning/checkpoints &&
gdown https://drive.google.com/uc?id=1TOJ3nsXnxfvMxygeXGQHxKH5-qOXeTBL &&
unzip ml_cascade_checkpoints.zip &&
rm -rf ml_cascade_checkpoints.zip &&
cd ../../

Download Example image

mkdir example_img &&
cd example_img &&
gdown https://drive.google.com/uc?id=1VYPWmU2QuEZ3Ys9LhAsDZp19dZmaaT4r &&
cd ../

Runing a full case

This terminal commands should be excecuted locally not in colab.

AIA-ML

python mc_detector.py --dcm-filepath <ABOSULTE_PATH_TO_REPO>/example_img/24065734_5291e1aee2bbf5df_MG_L_CC_ANON.dcm --detector-type 'aia_ml' --ouput-path /<ABOSULTE_PATH_TO_REPO>/example_img/ --store-csv --v

Deep learning classification based detection

python mc_detector.py --dcm-filepath <ABOSULTE_PATH_TO_REPO>/example_img/24065734_5291e1aee2bbf5df_MG_L_CC_ANON.dcm --detector-type 'classification_dl' --ouput-path /<ABOSULTE_PATH_TO_REPO>/example_img/ --store-csv --v --batch-size 224

Deep learning detection based detection

python mc_detector.py --dcm-filepath <ABOSULTE_PATH_TO_REPO>/example_img/24065734_5291e1aee2bbf5df_MG_L_CC_ANON.dcm --detector-type 'detection_dl' --ouput-path /<ABOSULTE_PATH_TO_REPO>/example_img/ --store-csv --v --batch-size 1

Further instructions

If the reader wants to run the full code, then downloading and preparation of the INBreast should be done as following:

Download and prepare INBreast database

cd data &&
gdown https://drive.google.com/uc?id=1ebw9N2vZY19TuELBZb39eAJhPjY1eFZX &&
unzip 'INbreast Release 1.0.zip' &&
rm -rf 'INbreast Release 1.0.zip' &&
cd ../ &&
python database/parsing_metadata.py --ib-path data/INbreast\ Release\ 1.0/ --rp --cb --pect-musc-mask

joaco18/calc-det