Robust Distance Estimation with Out-of-distribution Detection in Ophthalmic Surgery.

Area of Science:

• Biomedical Engineering
• Ophthalmology
• Medical Imaging

Background:

• Micrometer-scale precision is crucial for patient safety in ophthalmic surgery.
• Instrument-integrated optical sensors aim to measure instrument-to-tissue distances accurately.
• Segmentation errors from signal artifacts often compromise measurement reliability.

Purpose of the Study:

• To develop a deep learning framework for identifying and rectifying segmentation errors in optical coherence tomography (OCT) M-scans.
• To enhance the accuracy of instrument-to-retina distance estimation in ophthalmic procedures.
• To improve patient safety through more reliable surgical measurements.

Main Methods:

• A deep learning framework was developed to detect out-of-distribution OCT M-scans.
• The method incorporates adaptive remote center of motion (RCM)-informed retinal modeling and time series analysis.
• Retinal distances and confidence levels are estimated using retinal models, instrument positions, and validated data.

Main Results:

• The pipeline achieved 88.8% accuracy in identifying out-of-distribution measurements on ex vivo human eyes.
• Distance estimation accuracy improved by 89% and 93% compared to existing methods.
• A mean absolute error (MAE) of less than 40 μm was achieved across diverse conditions, including scans with blood and obstructions.

Conclusions:

• The proposed method significantly enhances the accuracy of instrument-to-retina distance estimation.
• This advancement contributes to improved patient safety in ophthalmic surgery.
• The framework shows potential for broader applications in other surgical disciplines with sensor-equipped instruments.