Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

BioTamperNet: Affinity-Guided State-Space Model Detecting Tampered Biomedical Images

Soumyaroop Nandi, Prem Natarajan

TL;DR

BioTamperNet tackles automated detection of duplicated manipulations in biomedical images by introducing affinity-guided self-attention and cross-attention built on lightweight State Space Model-inspired linear attention. The Siamese architecture jointly models intra- and inter-image correspondences to localize tampered regions and identify their sources, achieving state-of-the-art results on BioFors across External Duplication Detection, Internal Duplication Detection, and Cut/Sharp Transition Detection. The method is computationally efficient (36.7M parameters, 29.6 GFLOPs) and benefits from synthetic augmentation and GAN-based realism to bridge data scarcity in biomedicine. Together, these contributions advance reliable, cross-modality forensic tools for validating published biomedical research and enhancing scientific integrity.

Abstract

We propose BioTamperNet, a novel framework for detecting duplicated regions in tampered biomedical images, leveraging affinity-guided attention inspired by State Space Model (SSM) approximations. Existing forensic models, primarily trained on natural images, often underperform on biomedical data where subtle manipulations can compromise experimental validity. To address this, BioTamperNet introduces an affinity-guided self-attention module to capture intra-image similarities and an affinity-guided cross-attention module to model cross-image correspondences. Our design integrates lightweight SSM-inspired linear attention mechanisms to enable efficient, fine-grained localization. Trained end-to-end, BioTamperNet simultaneously identifies tampered regions and their source counterparts. Extensive experiments on the benchmark bio-forensic datasets demonstrate significant improvements over competitive baselines in accurately detecting duplicated regions. Code - https://github.com/SoumyaroopNandi/BioTamperNet

BioTamperNet: Affinity-Guided State-Space Model Detecting Tampered Biomedical Images

TL;DR

BioTamperNet tackles automated detection of duplicated manipulations in biomedical images by introducing affinity-guided self-attention and cross-attention built on lightweight State Space Model-inspired linear attention. The Siamese architecture jointly models intra- and inter-image correspondences to localize tampered regions and identify their sources, achieving state-of-the-art results on BioFors across External Duplication Detection, Internal Duplication Detection, and Cut/Sharp Transition Detection. The method is computationally efficient (36.7M parameters, 29.6 GFLOPs) and benefits from synthetic augmentation and GAN-based realism to bridge data scarcity in biomedicine. Together, these contributions advance reliable, cross-modality forensic tools for validating published biomedical research and enhancing scientific integrity.

Abstract

We propose BioTamperNet, a novel framework for detecting duplicated regions in tampered biomedical images, leveraging affinity-guided attention inspired by State Space Model (SSM) approximations. Existing forensic models, primarily trained on natural images, often underperform on biomedical data where subtle manipulations can compromise experimental validity. To address this, BioTamperNet introduces an affinity-guided self-attention module to capture intra-image similarities and an affinity-guided cross-attention module to model cross-image correspondences. Our design integrates lightweight SSM-inspired linear attention mechanisms to enable efficient, fine-grained localization. Trained end-to-end, BioTamperNet simultaneously identifies tampered regions and their source counterparts. Extensive experiments on the benchmark bio-forensic datasets demonstrate significant improvements over competitive baselines in accurately detecting duplicated regions. Code - https://github.com/SoumyaroopNandi/BioTamperNet
Paper Structure (23 sections, 27 equations, 11 figures, 5 tables)

This paper contains 23 sections, 27 equations, 11 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Training Dataset
  3. Proposed Method
  4. Preliminaries - State Space Models
  5. BioTamperNet Architecture Overview
  6. Siamese Duplication Detector Module
  7. Affinity Block.
  8. Affinity-Guided Self-Attention.
  9. Affinity-Guided Cross-Attention.
  10. Proposition 1 (Cross-View Duplication Alignment Under Affinity-Guided Attention).
  11. Siamese Decoder
  12. Training Details
  13. Experimental Evaluation
  14. Dataset and Metrics
  15. Results and Analysis
  16. ...and 8 more sections

Figures (11)

  • Figure 1: Microscopy, Blot/Gel, FACS, Macroscopy in BioFors sabir2021biofors.
  • Figure 2: EDD uses image pairs for detection; IDD and CSTD use single image each.
  • Figure 3: The EDD Synthetic Training Setup.
  • Figure 3: MCC for BioTamperNet EDD Ablation.
  • Figure 4: MCC for Cross-Modality Fine-Tuning Ablation on EDD.
  • ...and 6 more figures

Theorems & Definitions (1)

  • proof