User login
. AI detected more than 99% of all skin cancers.
The researchers tested the AI by integrating it into a clinical diagnosis process – anticipating a future in which AI helps doctors catch skin cancer faster and triage patients.
Skin cancer is the most common cancer in the United States; one in five 5 Americans develop skin cancer by age 70. With melanoma, the deadliest skin cancer, the 5-year survival rate is better than 99% if caught early, though only about three-quarters of melanomas are caught at this stage.
Amid rising skin cancer rates come concerns that the number of dermatologists in the workforce isn’t keeping pace. That may be why the average wait time for a dermatology appointment is trending up – in 2022, it reached 34.5 days.
The study, which was presented at the European Academy of Dermatology and Venereology Congress recently and has not yet been published, involved 6,900 patients in the United Kingdom with suspected skin cancer. The patients had been referred by their primary care physicians. The researchers took images of the suspicious areas and uploaded them to the AI software. The AI’s assessment was then shared with a dermatologist.
“Note that the diagnosis issued by the AI was not hidden from the dermatologist doing the second assessment,” said lead researcher Kashini Andrew, MBBS, a dermatologist and specialist registrar at University Hospitals Birmingham NHS Foundation Trust.
Dr. Andrew acknowledged that this may have influenced the dermatologist’s opinion. But that’s the vision of how doctors could use this tool.
The AI caught 59 of 59 melanomas and 189 of 190 total skin cancers (99.5%). (The one case that the AI missed was caught by the dermatologist.) It also flagged 541 of 585 precancerous lesions (92.5%). This represented a big improvement from a 2021 version of the model, which detected 86% of melanomas, 84% of all skin cancers, and 54% of precancerous lesions.
Over the 10-month period of the study, the system saved more than 1,000 face-to-face consultations, freeing dermatologists’ time to catch more cancers and serve more patients.
Limitations
The patients in the study were from “one hospital in a single region of the UK,” and the sample was not large enough to allow broad statements to be made about the use of AI in dermatology, Dr. Andrew said.
But it can open the conversation. Roxana Daneshjou, MD, PhD, a dermatologist at Stanford (Calif.) University who has studied the pros and cons of AI in medicine, had some concerns. For one thing, doctors can gather more in-depth information during an in-person exam than AI can glean from a photo, Dr. Daneshjou noted. They can examine skin texture, gather patient history, and take photos with special lighting and magnification.
And the AI needs to get better at ruling out malignancy, Dr. Daneshjou said. In this study, the AI identified 75% of benign lesions, a decline from the earlier version. The researchers noted in the abstract that this is a potential trade-off for increased sensitivity.
“[Unnecessary] biopsies can clog up the health care system, cost money, and cause stress and scarring,” said Dr. Daneshjou. “You don’t want to increase the burden of that.”
Still, if AI software such as the kind used in the study proves just as accurate in larger, more diverse sample sizes, then it could be a powerful tool for triage, Dr. Daneshjou said. “If AI gets particularly good at finding malignancy and also ruling it out, that would be a win.”
A version of this article appeared on Medscape.com.
. AI detected more than 99% of all skin cancers.
The researchers tested the AI by integrating it into a clinical diagnosis process – anticipating a future in which AI helps doctors catch skin cancer faster and triage patients.
Skin cancer is the most common cancer in the United States; one in five 5 Americans develop skin cancer by age 70. With melanoma, the deadliest skin cancer, the 5-year survival rate is better than 99% if caught early, though only about three-quarters of melanomas are caught at this stage.
Amid rising skin cancer rates come concerns that the number of dermatologists in the workforce isn’t keeping pace. That may be why the average wait time for a dermatology appointment is trending up – in 2022, it reached 34.5 days.
The study, which was presented at the European Academy of Dermatology and Venereology Congress recently and has not yet been published, involved 6,900 patients in the United Kingdom with suspected skin cancer. The patients had been referred by their primary care physicians. The researchers took images of the suspicious areas and uploaded them to the AI software. The AI’s assessment was then shared with a dermatologist.
“Note that the diagnosis issued by the AI was not hidden from the dermatologist doing the second assessment,” said lead researcher Kashini Andrew, MBBS, a dermatologist and specialist registrar at University Hospitals Birmingham NHS Foundation Trust.
Dr. Andrew acknowledged that this may have influenced the dermatologist’s opinion. But that’s the vision of how doctors could use this tool.
The AI caught 59 of 59 melanomas and 189 of 190 total skin cancers (99.5%). (The one case that the AI missed was caught by the dermatologist.) It also flagged 541 of 585 precancerous lesions (92.5%). This represented a big improvement from a 2021 version of the model, which detected 86% of melanomas, 84% of all skin cancers, and 54% of precancerous lesions.
Over the 10-month period of the study, the system saved more than 1,000 face-to-face consultations, freeing dermatologists’ time to catch more cancers and serve more patients.
Limitations
The patients in the study were from “one hospital in a single region of the UK,” and the sample was not large enough to allow broad statements to be made about the use of AI in dermatology, Dr. Andrew said.
But it can open the conversation. Roxana Daneshjou, MD, PhD, a dermatologist at Stanford (Calif.) University who has studied the pros and cons of AI in medicine, had some concerns. For one thing, doctors can gather more in-depth information during an in-person exam than AI can glean from a photo, Dr. Daneshjou noted. They can examine skin texture, gather patient history, and take photos with special lighting and magnification.
And the AI needs to get better at ruling out malignancy, Dr. Daneshjou said. In this study, the AI identified 75% of benign lesions, a decline from the earlier version. The researchers noted in the abstract that this is a potential trade-off for increased sensitivity.
“[Unnecessary] biopsies can clog up the health care system, cost money, and cause stress and scarring,” said Dr. Daneshjou. “You don’t want to increase the burden of that.”
Still, if AI software such as the kind used in the study proves just as accurate in larger, more diverse sample sizes, then it could be a powerful tool for triage, Dr. Daneshjou said. “If AI gets particularly good at finding malignancy and also ruling it out, that would be a win.”
A version of this article appeared on Medscape.com.
. AI detected more than 99% of all skin cancers.
The researchers tested the AI by integrating it into a clinical diagnosis process – anticipating a future in which AI helps doctors catch skin cancer faster and triage patients.
Skin cancer is the most common cancer in the United States; one in five 5 Americans develop skin cancer by age 70. With melanoma, the deadliest skin cancer, the 5-year survival rate is better than 99% if caught early, though only about three-quarters of melanomas are caught at this stage.
Amid rising skin cancer rates come concerns that the number of dermatologists in the workforce isn’t keeping pace. That may be why the average wait time for a dermatology appointment is trending up – in 2022, it reached 34.5 days.
The study, which was presented at the European Academy of Dermatology and Venereology Congress recently and has not yet been published, involved 6,900 patients in the United Kingdom with suspected skin cancer. The patients had been referred by their primary care physicians. The researchers took images of the suspicious areas and uploaded them to the AI software. The AI’s assessment was then shared with a dermatologist.
“Note that the diagnosis issued by the AI was not hidden from the dermatologist doing the second assessment,” said lead researcher Kashini Andrew, MBBS, a dermatologist and specialist registrar at University Hospitals Birmingham NHS Foundation Trust.
Dr. Andrew acknowledged that this may have influenced the dermatologist’s opinion. But that’s the vision of how doctors could use this tool.
The AI caught 59 of 59 melanomas and 189 of 190 total skin cancers (99.5%). (The one case that the AI missed was caught by the dermatologist.) It also flagged 541 of 585 precancerous lesions (92.5%). This represented a big improvement from a 2021 version of the model, which detected 86% of melanomas, 84% of all skin cancers, and 54% of precancerous lesions.
Over the 10-month period of the study, the system saved more than 1,000 face-to-face consultations, freeing dermatologists’ time to catch more cancers and serve more patients.
Limitations
The patients in the study were from “one hospital in a single region of the UK,” and the sample was not large enough to allow broad statements to be made about the use of AI in dermatology, Dr. Andrew said.
But it can open the conversation. Roxana Daneshjou, MD, PhD, a dermatologist at Stanford (Calif.) University who has studied the pros and cons of AI in medicine, had some concerns. For one thing, doctors can gather more in-depth information during an in-person exam than AI can glean from a photo, Dr. Daneshjou noted. They can examine skin texture, gather patient history, and take photos with special lighting and magnification.
And the AI needs to get better at ruling out malignancy, Dr. Daneshjou said. In this study, the AI identified 75% of benign lesions, a decline from the earlier version. The researchers noted in the abstract that this is a potential trade-off for increased sensitivity.
“[Unnecessary] biopsies can clog up the health care system, cost money, and cause stress and scarring,” said Dr. Daneshjou. “You don’t want to increase the burden of that.”
Still, if AI software such as the kind used in the study proves just as accurate in larger, more diverse sample sizes, then it could be a powerful tool for triage, Dr. Daneshjou said. “If AI gets particularly good at finding malignancy and also ruling it out, that would be a win.”
A version of this article appeared on Medscape.com.
FROM THE EADV CONGRESS