Generative Inpainting-Based Anomaly Detection for CT Liver Tumor Detection.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiology

Background:

• Computed Tomography (CT) is crucial for diagnosing liver diseases and tumors.
• Traditional anomaly detection in CT images can miss subtle tissue differences.
• Existing methods may yield suboptimal results in identifying pathological structures.

Purpose of the Study:

• To develop an advanced anomaly detection technique for liver CT imaging.
• To leverage generative diffusion models for improved accuracy in liver disease diagnosis.
• To refine the localization and detection of liver tumors using AI.

Main Methods:

• Employing a generative diffusion prior to inpaint liver CT images.
• Utilizing an adaptive threshold for abnormal region masking.
• Calculating anomaly scores based on image discrepancies post-inpainting.

Main Results:

• Demonstrated significant improvement in liver anomaly detection accuracy.
• Achieved a 7.9% increase in the area under the curve (AUC) compared to existing methods.
• Validated the methodology on two distinct liver CT datasets.

Conclusions:

• The generative diffusion prior approach enhances radiological assessment of liver diseases.
• This AI-driven method offers superior performance in detecting liver abnormalities.
• The technique shows promise for more precise and reliable liver tumor detection.