Artificial intelligence has come to the fore with a remarkable area of use in cancer research.
Researchers at Mass General Brigham in the US have developed two artificial intelligence tools that evaluate age-related features and survival risk by analyzing facial photographs of cancer patients. Introduced in the study published in The Lancet Digital Health Journal, FAHR-FaceAge estimates age from facial appearance, while FAHR-FaceSurvival creates a risk calculation regarding the survival risk of patients.
TRAINED WITH MORE THAN 40 MILLION PHOTOS
Stating that the basic artificial intelligence model was trained with more than 40 million facial photographs, Şendur said that the systems analyze complex patterns in different parts of the face.
Stating that the most important finding of the research is that the information obtained from facial photographs can carry meaningful clues about the general health status and survival risk of cancer patients, Şendur said, "The death risk of patients in the highest risk group was found to be approximately 3.23 times higher after other clinical factors were taken into account compared to the lowest risk group. However, this does not mean that how long a patient will live can be determined precisely from his photograph."
FACE AGE IS NOT THE SAME AS BIOLOGICAL AGE
Emphasizing that the FAHR-FaceAge system does not directly measure the biological age of organs, Şendur drew attention to the difference between chronological age and biological age. Stating that chronological age refers to the time since birth and biological age refers to the wear and tear and functional state of the body, Şendur said that the system produces an age-related estimate from facial appearance. Şendur stated that centennial age being higher than the calendar age is associated with poorer survival in some patient groups, and noted that this alone is not enough to change the chemotherapy dose or discontinue the treatment.
OVER 34 THOUSAND CANCER PATIENTS WERE EXAMINED
Stating that the survival model was adapted with photographs of 34 thousand 363 cancer patients, Şendur stated that the study is strong in terms of large data usage and validation in independent patient groups. However, Şendur said that the fact that the data were obtained mainly from certain health systems and especially from patient groups related to radiotherapy could make it difficult to generalize the results to all cancer types. Şendur also emphasized that it has not yet been proven that an algorithm that is successful in retrospective evaluations will provide the same benefit in the real clinical environment.
THE TREATMENT PROCESS MAY CHANGE THE FACIAL APPEARANCE
Pointing out that cancer itself and the treatments applied can affect facial appearance, Şendur stated that weight and muscle loss, dehydration, edema, corticosteroid use, sleep disorders and nutritional status can affect artificial intelligence predictions. "Some of these changes may be a sign of the general vulnerability of the patient, but they do not always reflect the biological aggressiveness of the cancer," said Şendur, adding that the time when the photo was taken, the treatment period and clinical conditions should be standardized.
MUST BE VERIFIED BEFORE USE IN Türkiye
Emphasizing that the models should not be considered as independent tests that determine routine treatment decisions for now, Şendur stated that treatment decisions in oncology are based on many factors such as the type and stage of cancer, molecular features, the patient's performance status and accompanying diseases.
Stating that the algorithms should also be examined with the same accuracy in different skin tones, ethnicities and age groups, Şendur said that prospective validation studies involving participants from different centers and patient groups are needed for use in Türkiye. Şendur concluded his words by stating that photographing conditions should be standardized, the results should be compared with clinical data, and data security and personal data protection should be ensured:
"The role of artificial intelligence should not be to replace the doctor or the patient-physician relationship, but to provide verified, transparent and safe decision support. The right approach for Turkey is first local scientific validation, then controlled clinical use."
