Figure 3.

Quality issues in the smile task. Common problems included participants being too far from the camera, which limited the visibility of subtle facial movements for both human raters and AI-based pose tracking. Glare from eyeglasses often obscured the eyes, preventing reliable eye tracking and blink detection—important markers in movement disorder assessment. Another frequent issue was the presence of multiple people in the frame, which complicates automated analysis by requiring subject identification. The middle image represents real data but was generated by ChatGPT 5 [77] to preserve participant anonymity using the prompt “Generate an image of a 60-year-old male. The man is bald, wearing a black t-shirt, and sitting in a chair at his house. The background is a slightly dark living room (with windows, hanging photos, and other furniture). The man is wearing eyeglasses, and bright light is reflected from them. His eyes are severely obstructed due to these reflections.