Deep learning for detection of iso-dense, obscure masses in mammographically dense breasts

Krithika Rangarajan*1,2, Pranjal Aggarwal*1, Dhruv Kumar Gupta1, Rohan Dhanakshirur1, Akhil Baby2, Chandan Pal2, Arun Kumar Gupta2, Smriti Hari2, Subhashis Banerjee1, Chetan Arora1
1Indian Institute of Technology Delhi
2All India Institute of Medical Sciences, Delhi
*Equal Contribution
Paper Demo Code

We propose a novel deep-learning based method for incorporating traditional radiology principles to improve cancer detection in mammographically dense breasts.

Summary of our Study Methodology.

Abstract

We present a deep learning model designed to address the challenges of detecting isodense and obscure masses in mammographically dense breasts. By incorporating traditional radiology principles our model demonstrates improved sensitivity in detecting malignancies compared to baseline networks. We showcase the performance of our method on diagnostic and screening mammography datasets, highlighting its potential to enhance cancer detection accuracy in dense breasts and its applicability to different patient distributions.

Key Highlights 🌟

  • 🎯Addresses the challenge of detecting isodense, obscure masses and mammographically dense breasts in deep learning networks.
  • 🤝Collaborative network design and incorporation of traditional radiology teaching improves cancer detection in dense breasts.
  • 📊Demonstrates increased sensitivity for malignancy in diagnostic mammography dataset and external validation test set with a screening mammography distribution.
  • 🌐Our method's accuracy is translatable to different patient distributions, showcasing its potential in both screening and diagnostic mammography datasets.
  • 🩺Clinical relevance: Incorporating medical knowledge into neural network design can help overcome limitations associated with specific modalities, improving performance on mammographically dense breasts.

Results Summary

We evaluate our method across different settings and datasets.
FROC curve on Diagnostic Mammography Dataset for different settings
FROC curve on Screening Mammography Dataset for different settings
Performance on public benchmark InBreast
Component Wise Analysis of Our Method

Citation

TY  - JOUR
      AU  - Rangarajan, Krithika
      AU  - Agarwal, Pranjal
      AU  - Gupta, Dhruv Kumar
      AU  - Dhanakshirur, Rohan
      AU  - Baby, Akhil
      AU  - Pal, Chandan
      AU  - Gupta, Arun Kumar
      AU  - Hari, Smriti
      AU  - Banerjee, Subhashis
      AU  - Arora, Chetan
      PY  - 2023
      DA  - 2023/05/20
      TI  - Deep learning for detection of iso-dense, obscure masses in mammographically dense breasts
      JO  - European Radiology
      AB  - To analyze the performance of deep learning in isodense/obscure masses in dense breasts. To build and validate a deep learning (DL) model using core radiology principles and analyze its performance in isodense/obscure masses. To show performance on screening mammography as well as diagnostic mammography distribution.
      SN  - 1432-1084
      UR  - https://doi.org/10.1007/s00330-023-09717-7
      DO  - 10.1007/s00330-023-09717-7
      ID  - Rangarajan2023
      ER  -