Skip to content

Latest commit

 

History

History
36 lines (30 loc) · 695 Bytes

README.md

File metadata and controls

36 lines (30 loc) · 695 Bytes

EngineerDocs

Teknofest AI in Healthcare Category Finalists

This repositorty contains the notebooks used to Preprocess and Train

Used Datasets:

  • VinDr
  • RSNA
  • Teknofest Data

Tried models:

  • MobileNetv2
  • Inception v3
  • VGG
  • MNasNet
  • ConvNeXt
  • EfficientNet models
  • ViT models
  • Dyno v2
  • LightGBM
  • XGBoost

EfficientNet_b3 was used as the finalized model. FocalLoss is implemented to learn on non-equally distributed multiclass dataset. Google Colab is used for training hardware supplier. The project report is also uploaded in the repository.

Targets:

BIRADS Category

  • BIRADS 0
  • BIRADS 1-2
  • BIRADS 4-5

Breast Density (Composition)

  • A
  • B
  • C
  • D