Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Unsupervised Hybrid framework for ANomaly Detection (HAND) -- applied to Screening Mammogram

Zhemin Zhang, Bhavika Patel, Bhavik Patel, Imon Banerjee

TL;DR

The proposed HAND pipeline offers an automated efficient computational solution for domain-specific quality checks in external screening mammograms, yielding actionable insights without direct exposure to the private medical imaging data.

Abstract

Out-of-distribution (OOD) detection is crucial for enhancing the generalization of AI models used in mammogram screening. Given the challenge of limited prior knowledge about OOD samples in external datasets, unsupervised generative learning is a preferable solution which trains the model to discern the normal characteristics of in-distribution (ID) data. The hypothesis is that during inference, the model aims to reconstruct ID samples accurately, while OOD samples exhibit poorer reconstruction due to their divergence from normality. Inspired by state-of-the-art (SOTA) hybrid architectures combining CNNs and transformers, we developed a novel backbone - HAND, for detecting OOD from large-scale digital screening mammogram studies. To boost the learning efficiency, we incorporated synthetic OOD samples and a parallel discriminator in the latent space to distinguish between ID and OOD samples. Gradient reversal to the OOD reconstruction loss penalizes the model for learning OOD reconstructions. An anomaly score is computed by weighting the reconstruction and discriminator loss. On internal RSNA mammogram held-out test and external Mayo clinic hand-curated dataset, the proposed HAND model outperformed encoder-based and GAN-based baselines, and interestingly, it also outperformed the hybrid CNN+transformer baselines. Therefore, the proposed HAND pipeline offers an automated efficient computational solution for domain-specific quality checks in external screening mammograms, yielding actionable insights without direct exposure to the private medical imaging data.

Unsupervised Hybrid framework for ANomaly Detection (HAND) -- applied to Screening Mammogram

TL;DR

The proposed HAND pipeline offers an automated efficient computational solution for domain-specific quality checks in external screening mammograms, yielding actionable insights without direct exposure to the private medical imaging data.

Abstract

Out-of-distribution (OOD) detection is crucial for enhancing the generalization of AI models used in mammogram screening. Given the challenge of limited prior knowledge about OOD samples in external datasets, unsupervised generative learning is a preferable solution which trains the model to discern the normal characteristics of in-distribution (ID) data. The hypothesis is that during inference, the model aims to reconstruct ID samples accurately, while OOD samples exhibit poorer reconstruction due to their divergence from normality. Inspired by state-of-the-art (SOTA) hybrid architectures combining CNNs and transformers, we developed a novel backbone - HAND, for detecting OOD from large-scale digital screening mammogram studies. To boost the learning efficiency, we incorporated synthetic OOD samples and a parallel discriminator in the latent space to distinguish between ID and OOD samples. Gradient reversal to the OOD reconstruction loss penalizes the model for learning OOD reconstructions. An anomaly score is computed by weighting the reconstruction and discriminator loss. On internal RSNA mammogram held-out test and external Mayo clinic hand-curated dataset, the proposed HAND model outperformed encoder-based and GAN-based baselines, and interestingly, it also outperformed the hybrid CNN+transformer baselines. Therefore, the proposed HAND pipeline offers an automated efficient computational solution for domain-specific quality checks in external screening mammograms, yielding actionable insights without direct exposure to the private medical imaging data.
Paper Structure (15 sections, 4 equations, 4 figures, 8 tables)

This paper contains 15 sections, 4 equations, 4 figures, 8 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Method
  4. Synthetic OOD Generation for SSL training
  5. Hybrid Model Architecture
  6. Anomaly detection
  7. Results
  8. Datasets
  9. Baselines
  10. Quantitative Performance
  11. Qualitative Performance
  12. Ablation Study
  13. Discussion
  14. HAND Network Details
  15. Data Pre-processing

Figures (4)

  • Figure 1: Comparative examples between ID, intra-class and inter-class - (a) ID mammogram. (b) Intra-class OOD example is biopsy clip and (c) is implant rapture, which share significant similarity with ID mammogram. (d) Inter-class OOD example from ImageNet, which is significantly different from mammogram images.
  • Figure 2: HAND: Proposed hybrid model framework with synthetic ODD generation: A) model architecture by combining CNN and transformer backbone; B) six different transformations for synthetic OOD generation.
  • Figure 3: Comparative analysis of various models - (a) SSIM values for different models for ID, ODD and natural image data; (b) OOD classification performance reported in terms of AUC on the internal and external dataset.
  • Figure 4: Examples of pre-process mammograms (crop and stitch right and left MLO view under same study ID) - (a) Original MLO right view. (b) Original MLO left view. (c) Cropped and stitched MLO right and MLO left under same study ID.