Labyrinth net: A robust segmentation method for inner ear labyrinth in CT images.

Area of Science:

• Medical Imaging
• Computer-Aided Diagnosis
• Otology

Background:

• The inner ear labyrinth is crucial for hearing and balance, located within the petrous temporal bone.
• Accurate segmentation of the inner ear labyrinth is vital for quantitative measurements and diagnosing ear diseases.
• Deep learning segmentation faces challenges due to the labyrinth's complex morphology, small size, and high labeling costs.

Purpose of the Study:

• To develop a robust and efficient automatic segmentation method for the inner ear labyrinth in temporal bone CT images.
• To reduce the cost and improve the efficiency of data labeling for deep learning models.

Main Methods:

• Proposed a multi-model inconsistency approach within an active learning paradigm.
• Introduced an informative sample assessment strategy using an observer network for confidence assessment.
• Developed a maximum-connected probability map (MCP-Map) to refine segmentation by reducing outlier influence.

Main Results:

• The proposed method demonstrated the highest labeling efficiency and lowest labeling cost compared to existing active learning techniques.
• Achieved 95.67% Dice Similarity Coefficient (DSC) with a 40% reduction in labeled data.
• Represents a state-of-the-art performance in labyrinth segmentation.

Conclusions:

• The developed method offers a highly efficient and accurate solution for segmenting the inner ear labyrinth.
• This approach significantly lowers the barrier for applying deep learning in otology research and clinical diagnosis.
• The technique holds promise for advancing computer-aided diagnosis in ear diseases.