Transdermal optical imaging, using machine learning and video from a smartphone camera and using advanced machine learning may^[2][3] or may not^[4] be able to determine a subject’s blood pressure.

Imaging may also be able to detect:

• Heart rate^[5][3]
• Atrial fibrillation^[6]
• Blood pressure^[2]. The goal for accuracy is the ISO standard of “a device is considered acceptable if its estimated probability of a tolerable error (≤10 mmHg) is at least 85%”^[7] or “average difference no greater than 5 mmHg and SD no greater than 8 mmHg”^[8].
• Respiratory rate^[9]
• Jaundice^[10]
• Ethanol intoxication^[11].
• Cushing’s syndrome^[12]
• Acromegaly^[12]

Imaging of the retina, using deep-learning trained on data from 284,335 patients, may predict^[13]:

• age (mean absolute error within 3.26 years)
• gender (area under the receiver operating characteristic curve (AUC) = 0.97)
• smoking status (AUC = 0.71)
• systolic blood pressure (mean absolute error within 11.23 mmHg)
• major adverse cardiac events (AUC = 0.70)

Retina imaging with deep learning can detect papilledema^[14].

Pain sensitivity has been measured^[15].

Thermal image sensors may be helpful^[16].

Legal issues have been debated about the role of transparency and human oversight in interpreting information derived from deep learning^[17][18].

The ability to generalize the accuracy of image analyses to images different that those that trained the system may be limited^[19].