Artifact Suppression

Identifies and suppresses linear, high-density curvilinear structures caused by improper breast positioning. Skin folds mimic the radiological appearance of architectural distortion, a key sign of malignancy. Algorithms analyze gradient orientation coherence and continuity across the breast boundary to distinguish superficial folds from parenchymal distortions. Suppression prevents these common positioning artifacts from generating false positive marks in the Region of Interest (ROI) detector.

Corrects for blur induced by patient movement during exposure. Motion unsharpness obscures fine microcalcification margins and can create spurious density gradients. Deconvolution algorithms estimate the point spread function (PSF) from the image's power spectrum and apply inverse filtering to restore edge sharpness. This ensures that true high-frequency lesion details are preserved for downstream patch-based analysis while preventing blur artifacts from being misinterpreted as ill-defined mass margins.

Segments and digitally removes radiopaque biopsy clips, surgical staples, and skin markers. These high-attenuation objects cause streak artifacts and photon starvation in Full-Field Digital Mammography (FFDM) and Digital Breast Tomosynthesis (DBT). A threshold-based segmentation combined with morphological inpainting replaces the metallic region with a texture-synthesized background. This prevents the detector from fixating on obvious foreign bodies, allowing it to focus on genuine microcalcification clusters.

Suppresses periodic line patterns introduced by the anti-scatter grid. These high-frequency, oriented lines can interfere with Fourier domain feature extractors in convolutional neural networks. A notch filter applied in the frequency domain targets the specific spatial frequency of the grid pattern. Removing this structured noise is critical for accurate radiomics feature extraction, as grid lines can corrupt texture-based biomarkers like co-occurrence matrices.

Mitigates the bright, sharp boundary artifact at the implant-tissue interface. The extreme density gradient at the edge of a silicone or saline implant generates a halo artifact that can be mistaken for a spiculated mass. Algorithms use prior knowledge of implant shape and a contrast normalization step to dampen this edge response without affecting the contrast of true lesions within the surrounding fibroglandular tissue.

Corrects for fixed-pattern noise caused by detector imperfections or debris on the imaging plate. These artifacts appear as consistent, high-contrast points across multiple exams. By maintaining a flat-field calibration map, the system subtracts known detector defects. This prevents a static dust speck from being algorithmically linked across temporal comparison studies and flagged as a developing interval cancer, a critical requirement for maintaining low recall rates.

A post-processing AI technique designed to suppress erroneous marks generated by a detection model, thereby improving specificity and reducing unnecessary recall rates. Artifact suppression is a critical pre-processing step for this, as it removes the non-anatomical signals that often trigger these false positives.

• Mechanism: Often uses a secondary convolutional neural network (CNN) trained on false-positive examples.
• Clinical Impact: Directly lowers the patient anxiety and healthcare costs associated with unnecessary diagnostic workups.
• Relationship: Artifact suppression is a primary input sanitization step; false positive reduction is the output filtering step.

An algorithmic process that geometrically links findings across the Craniocaudal (CC) and Mediolateral Oblique (MLO) views to confirm a true lesion and reduce false positives. Artifact suppression ensures that a skin fold on one view is not incorrectly correlated with a true mass on the other.

• Geometric Matching: Uses epipolar geometry to map a candidate region in one view to a corresponding band in the orthogonal view.
• Artifact Rejection: A true lesion will appear in both views, whereas a metallic marker or motion blur artifact typically will not, allowing the system to suppress the artifact.

The spatial alignment of a current mammogram with a historical one using rigid or deformable transformations to enable accurate side-by-side comparison. Artifact suppression is vital here to prevent the registration algorithm from locking onto a stable, non-anatomical marker (like a surgical clip) instead of the parenchymal tissue.

• Temporal Subtraction: After registration, the prior image is digitally subtracted from the current one to highlight subtle interval changes.
• Artifact Masking: Suppressed artifacts are masked out to prevent them from appearing as high-contrast 'changes' in the subtraction image.

A deep learning strategy where a large mammogram is divided into smaller, overlapping sub-images for high-resolution feature extraction before global aggregation. Artifact suppression algorithms often operate at this patch level to identify and normalize local anomalies.

• Local Context: A patch containing a skin fold can be identified by its characteristic linear, low-intensity pattern and suppressed before being passed to the detection network.
• Computational Efficiency: Suppressing artifact-heavy patches early in the pipeline prevents the downstream model from wasting compute on non-diagnostic information.

The automated assignment of an ACR density category (A through D) based on the ratio of fibroglandular tissue to adipose tissue. Dense tissue appears as bright, fibrous regions on a mammogram and can mimic or obscure suspicious lesions. Artifact suppression must be robust enough to distinguish true dense parenchyma from motion blur, which can have a similar textural appearance.

• Masking Risk: In dense breasts (Categories C and D), artifacts can further complicate an already challenging visual field.
• Algorithmic Sensitivity: Suppression algorithms must be tuned to avoid inadvertently removing the subtle architectural distortions that are often the only sign of cancer in dense tissue.

The handling, parsing, and interoperability of the Digital Imaging and Communications in Medicine standard. Artifact suppression metadata, such as the type of artifact removed and the region of the image that was altered, must be logged in the DICOM header to maintain a complete audit trail for regulatory compliance.

• Presentation State: A DICOM Grayscale Softcopy Presentation State (GSPS) object can store the parameters of the suppression operation without altering the original 'for processing' pixel data.
• Burned-In Annotation: If the suppression results in a permanent alteration of the image for display, this must be flagged in the DICOM tags to inform the interpreting radiologist.