Reverse image search often yields wrong diagnosis for skin disease

News Hour:

Google’s reverse image search tool isn’t designed to diagnose skin conditions, researchers say, and people who try to use it this way are likely to get a wrong result.

But in a test using photographs of benign and cancerous skin conditions, the top 10 matches returned were often not the same disease at all.

“As expected, the Google reverse search engine does a great job of recognizing objects such as houses, refrigerators, animals,” study coauthor Dr. Kavita Y. Sarin told Reuters Health by email. “When it comes to skin lesions, the accuracy drops significantly,” said Sarin, of the dermatology department at Stanford University School of Medicine in Redwood City, California, reports Reuters.

“Physicians undergo many years of training to diagnose many of these skin conditions and, when unsure, rely on additional tools such as dermoscopy and biopsy,” she said. “It is important for users to be aware of the limitations of relying on search engine alone for diagnosis.”

At Google Images (images.google.com), users can upload an image, or link to an image, and search for matching images, instead of using text search terms.

The researchers used the tool on 100 photographs of the 10 most common benign or malignant skin conditions from a database of U.S. Navy Medical Center images takes by board-certified dermatologists. These included pictures of cancers like melanoma and squamous cell carcinoma, and benign lesions like hemangioma and nevus spilus.

They measured the accuracy of the search by how many of the top 10 unfiltered image results displayed the same condition, which would represent a “correct diagnosis.”

Up to 30 percent of the time, there were no correct results in the top 10 images for skin cancer searches, and the error rate ranged from 30 to 100 percent for benign conditions, according to the report in JAMA Dermatology.

Adding the word “skin” as a text cue helped the image search tool perform better, and the error rate was smaller for malignant conditions like skin cancers, the researchers note.

People who use image search for diagnosis risk having a misdiagnosis and feeling inappropriately reassured, so they might not seek care, Sarin said. People should rely on their doctors for an accurate diagnosis.

“Patients will inevitably adopt new technologies to help understand their condition and answer health concerns,” Sarin said. “These technologies can equip patients with information and be learning tools but patients should understand that these technologies still are not very accurate and they should still consult their physician if they are concerned.”

For now, she recommends patients only use the Google image search tool for information or exploration but not to diagnose skin conditions.

Google declined to comment on the study.

Md. Rafiuzzaman Sifat, a CSE graduate turned into journalist, works at News Hour as a staff reporter. He has many years of experience in featured writing in different Bangladeshi newspapers. He is an active blogger, story writer and social network activist. He published a book named 'Se Amar Gopon' inEkushe boi mela Dhaka 2016. Sifat got a BSc. from Ahsanullah University of Science & Technology, Bangladesh. He also works as an Engineer at Bangla Trac Communications Ltd. As an avid traveler and a gourmet food aficionado, he is active in publishing restaurant reviews and cutting-edge articles about culinary culture.
No Comments