Written By:
Chris Cole
Managing Editor


Studies suggest that an accurate diagnosis of skin lesions is achieved predominately through non-analytical pattern recognition using training examples rather than by rule-based algorithms. “Melanoma is easily curable if recognized early,” says Michael S. Kolodney, MD, PhD. “Dermatologists are good at spotting melanomas because they develop an innate sense of how they appear after examining thousands of malignant and benign lesions. Conversely, most medical students are relatively disadvantaged by their limited dermatology exposure. Too little experience, rather than lack of knowledge of the rules, is the primary barrier to developing pattern-recognition and intuition as reliable tools for a melanoma diagnosis by non-experts.”

To overcome this issue, Dr. Kolodney and colleagues developed a novel web-based application that mimics the training of dermatologists by teaching medical students how to intuitively diagnose melanoma in a condensed timeframe. “Our application, called Skinder, teaches intuitive visual diagnosis of melanoma by quickly presenting learners with thousands of benign and malignant skin lesions. Users make rapid binary decisions by swiping right for benign or left for malignant, and then receive instant feedback on accuracy. With this application, learners can amass a mental repository of diagnostic experience in a short amount of time.”


A Closer Look

For a study published in JAMA Dermatology, Dr. Kolodney and colleagues compared Skinder with publicly available data from the Internet Curriculum for Melanoma Early Detection (INFORMED) Skin Education Series. A total of 36 medical students who did not have a formal clinical dermatology rotation were tested on their ability to differentiate melanomas from benign pigmented lesions before and after training with either Skinder or the INFORMED Skin Education Series. Participants were randomized into the rule-based or application groups, and each took a 32-image pretest in which they were asked to determine if lesions were a melanoma or a benign skin lesion. They were then given 60 minutes of observed training time devoted to their assigned learning modality. Immediately following this training, students took a posttest with the same 32 images in randomized order to evaluate improvement.

“Our study found that the average pretest score for medical students randomized to the Skinder group was 75.0% correct, compared with 74.7% correct for those randomized to the INFORMED group,” Dr. Kolodney says. “The posttest average score for the application group was 86.3% correct, compared with 77.5% correct for the INFORMED group, a finding that was highly significant.” During the 60-minute training session, application users estimated that Skinder held their attention for 34.3 minutes, which was less than the 45.6 minutes estimated by INFORMED trainees. Overall, Skinder users believed they were more likely than INFORMED users to access their learning modality again if given the opportunity (Table).


Assessing Implications

Despite the small sample size, the study by Dr. Kolodney and colleagues suggests that Skinder is a more effective learning tool for accurately diagnosing melanoma than traditional rule-based methods. The study also reinforces the importance of visual pattern recognition when clinically diagnosing melanoma and supports the premise that intuitive diagnoses are superior to rule-based algorithms. Future studies will expand the sample size of the current analysis and test whether the application is useful to primary care physicians and nurses and perhaps to patients who are at risk for melanoma.

“As students, we’re often taught to make visual diagnostic decisions based on rules, but in the real world, we often use a more intuitive approach based on experience,” says Dr. Kolodney. “Teaching students intuitive approaches may improve their visual diagnostic skills. There is no way to become an expert in something without experience. However, using technology to simulate clinical experience through smartphones apps is a promising approach to accelerate the learning curve.”