Missing-view completion for fatty liver disease detection.

Area of Science:

• Medical imaging and diagnostics
• Artificial intelligence in healthcare
• Noninvasive disease detection

Background:

• Fatty liver disease affects 10-30% of the population globally, potentially progressing to cirrhosis, liver failure, and cancer.
• Current noninvasive fatty liver disease detection relies on medical imaging (CT, ultrasound) and expert experience.
• Existing multi-view diagnostic methods struggle with incomplete or missing data, hindering accurate classification.

Purpose of the Study:

• To develop an intelligent electronic noninvasive diagnostic approach for fatty liver disease.
• To address the challenge of missing-view data in multi-view noninvasive disease detection.
• To improve the classification performance of fatty liver disease using multi-view data completion.

Main Methods:

• A novel multi-view completion method was proposed to process incomplete multi-view datasets.
• The method utilizes facial, sublingual vein, and tongue images to generate features for diagnostic evaluation.
• Intra-view and inter-view information were explored to effectively generate missing-view data.

Main Results:

• The proposed multi-view completion method demonstrated improved diagnostic results for fatty liver disease.
• Performance was enhanced compared to using original incomplete multi-view data across various classifiers.
• Experiments were conducted on a dataset of 220 fatty liver patients and 220 healthy controls.

Conclusions:

• The developed method effectively addresses the missing-view issue in noninvasive disease detection.
• Multi-view completion significantly improves the performance of computerized noninvasive diagnostic methods.
• This approach offers a promising advancement for early and accurate fatty liver disease diagnosis.