AI enhanced imaging: How shape models support endoscopic OCT imaging of the middle ear

Introduction Optical coherence tomography (OCT) allows a contact-free, high-resolution three-dimensional and functional imaging of tissues. Using endoscopic optics, OCT enables imaging of the tympanic membrane and adjacent structures. However, due to shadowing effects, parts of the ossicles are hidden, which complicates the interpretation of the resulting incomplete 3D data.

Methods An endoscopic OCT system with a field of view of 10 mm was used to obtain volume scans of 10 healthy subjects, which subsequently were manually segmented. Using these data, a neural network was trained for automated segmentation and labeling of the middle ear structures. A statistical shape model based on 50 μCT images of human temporal bones was fitted to the OCT volume data.

Results The use of the neural network resulted in reliable segmentation of the tympanic membrane, malleus handle, long incus process, and parts of the stapes superstructure. The use of the statistical shape model allowed better delineation of the depth extent of the ossicles.

Conclusion Endoscopic OCT shows high potential for middle ear diagnosis. Shortcomings due to missing image information can be compensated to a certain extent by a priori knowledge of the anatomy. With the approach presented here, further combination of image data, for example with radiological datasets, is possible.

Else-Kröner-Fresenius-Zentrum für Digital Health

Publication History

Article published online:
12 May 2023