/Breast-Cancer-Detection

Breast cancer detection with screening mammograms obtained from regular screening.

Primary LanguageJupyter NotebookMIT LicenseMIT

Breast-Cancer-Detection

Breast cancer detection with screening mammograms obtained from regular screening.

Motivation

According to the WHO, breast cancer is the most commonly occurring cancer worldwide. In 2020 alone, there were 2.3 million new breast cancer diagnoses and 685,000 deaths. Yet breast cancer mortality in high-income countries has dropped by 40% since the 1980s when health authorities implemented regular mammography screening in age groups considered at risk. Early detection and treatment are critical to reducing cancer fatalities. In this project, I developed multiple cancer detection models from routine mammography scans and compared their performance.

Analysis

Target column distribution

alt text

Correlation

alt text

Healthy breasts

alt text

Abnormal breasts

alt text

Natural breast with cancer

alt text

Implant breast with cancer

alt text

My Approch:

  • The given DCM image files were converted to PNG. I tried converting the file in the data generator, but it took extra time.
  • Most of the regions in the image didn't contain useful information. So, I extracted the Region of Interest (ROI) and created a new dataset to work with.
  • Transfer learning was used to evaluate the performance of various well-known architectures (AlexNet, VGG-16, Resnet50, and EfficientNet). The performance of various vision transformers was then evaluated.
  • The performance of vision transformer was satisfactory and the results are given in the table below.

    • Result

    Algorithm Accuracy AUC Pf1
    EfficientNetV1B0 0.9767 0.5920 0.0417
    ConvNeXtTiny 0.9793 0.7009 0.0647
    FlexiViTSmall 0.9778 0.5297 0.0305
    DaViT_T 0.9781 0.6953 0.0719

    ConvNeXtTiny history

    alt text