BLS-GAN: A Deep Layer Separation Framework for Eliminating Bone Overlap in Conventional Radiographs

TL;DR

This work tackles bone overlap in conventional radiographs by introducing BLS-GAN, a three-component framework that separates overlapped bone layers via a Layer Image Generator, a segmentation-based multi-channel Supervisor, and a Radiography Principles–based Reconstructor, with synthetic pre-training to stabilize learning. The approach adheres to radiographic physics while enabling independent analysis of each bone layer, demonstrated on MCP joints in RA patients. Empirical results show improved image realism and reconstruction quality (lower MSE and FID, higher SSIM/PSNR) and improved JSN quantification accuracy and stability, validated by expert assessments and clinical metrics. The method offers a practical path to extend automated MSK analysis to complex joints and supports future refinement of soft-tissue separation for broader radiographic applications.

Abstract

Conventional radiography is the widely used imaging technology in diagnosing, monitoring, and prognosticating musculoskeletal (MSK) diseases because of its easy availability, versatility, and cost-effectiveness. In conventional radiographs, bone overlaps are prevalent, and can impede the accurate assessment of bone characteristics by radiologists or algorithms, posing significant challenges to conventional and computer-aided diagnoses. This work initiated the study of a challenging scenario - bone layer separation in conventional radiographs, in which separate overlapped bone regions enable the independent assessment of the bone characteristics of each bone layer and lay the groundwork for MSK disease diagnosis and its automation. This work proposed a Bone Layer Separation GAN (BLS-GAN) framework that can produce high-quality bone layer images with reasonable bone characteristics and texture. This framework introduced a reconstructor based on conventional radiography imaging principles, which achieved efficient reconstruction and mitigates the recurrent calculations and training instability issues caused by soft tissue in the overlapped regions. Additionally, pre-training with synthetic images was implemented to enhance the stability of both the training process and the results. The generated images passed the visual Turing test, and improved performance in downstream tasks. This work affirms the feasibility of extracting bone layer images from conventional radiographs, which holds promise for leveraging bone layer separation technology to facilitate more comprehensive analytical research in MSK diagnosis, monitoring, and prognosis. Code and dataset: https://github.com/pokeblow/BLS-GAN.git.

Figures (7)

• Figure 1: Explanation of bone overlap: Due to finger flexion and the excessively narrow joint space, the normal joint highlighted in the yellow box in (A) appears as the joint with bone overlap in (B). Bone overlap impedes clinical imaging diagnosis and its automatic analysis in MSK diseases.
• Figure 2: Explanation of bone layer separation: Layer-by-layer extraction of the upper and lower bones, followed by eliminating overlapped regions. The framework consists of three primary components: a generator, a supervisor, and a reconstructor. This process is performed as follows: (i) The generator produces bone layer images using the original joint image and corresponding bone masks as input. (ii) The layer images are discriminated by a segmentation-based multi-channel network and reconstructed through reconstructor, yielding a discrimination mask and a reconstructed image. (iii) Discrepancies between the masks and ground truth (GT), and between the reconstructed and original images, are used to create a hybrid loss function that guides the generator and reconstructor during back propagation. (iv) In the training pipeline, pre-training is preformed in synthetic images, the discrepancies of the real bone layer images is incorporated into the original loss function to facilitate the establishment of the initial model. Subsequently, the training is performed in real and synthetic images.
• Figure 3: Comparison of proposed framework with other methods in different metrics across overlap sizes.
• Figure 4: Comparison of proposed framework with other methods. (A) Real Joint image; (B) Reconstructed Joint Image; (C) Upper Bone Layer; (D) Upper Bone Layer; (E) MSE Spectrum (A v.s. B).
• Figure 5: Ablation study results of our framework in different metrics across overlap sizes.