This work presents a clustering approach for Time-domain Full-field optical coherence tomography (TD-FFOCT) retinal images.TD-FFOCT is an efficient method for cellular-level analysis of retinal structures, with fast acquisition and wide field-of-view. However, clinical use faces challenges from involuntary axial retinal motion due to breathing, heartbeat, and pulsation. Despite real-time axial motion compensation, achieved precision is around 10µm rms, below the ideal 4µm for an 8µm coherence gate, impacting system robustness and image depth selection. One way to overcome this is to group together images featuring similar retinal structures and acquired at the same depth. We propose a comprehensive clustering approach using learning-based and non-learning-based methods for feature extraction and clustering. Results show that clustering can help mitigate the effects of motion on the acquired image data, improving imaging accuracy and robustness.
|