In 2016, more than 76,380 new melanoma cases were diagnosed and 10,130 people were expected to die from skin cancer in the United States (one death per hour) [1]. A recent study demonstrates that the economic burden of skin cancer treatment is substantial and, in the United States, the cost was increased from $3.6 billion in 2002–2006 to $8.1 billion in 2007–2011 [2].
Monitoring moderate and high-risk patients and identifying melanoma in the earliest stage of disease should save lives and greatly diminish the cost of treatment. In this project, we are focused on detection and monitoring of new potential melanoma sites with medium/high risk patients. We believe those patients have a serious need and they need to be motivated to be engaged in their treatment plan. High-risk patients are more likely to be engaged with their skin health and their health care providers (physicians). Considering the high morbidity and mortality of melanoma, these patients are motivated to spend money on low-cost mobile device technology, either from their own pocket or through their health care provider if it helps reduce their risk with early detection and treatment. We believe that there is a role for mobile device imaging tools in the management of melanoma risk, if they are based on clinically validated technology that supports the existing needs of patients and the health care system.
In a study issued in the British Journal of Dermatology [2] of 39 melanoma apps [2], five requested to do risk assessment, while nine mentioned images for expert review. The rest fell into the documentation and education categories. This seems like to be reliable with other dermatology apps available on the market. In a study at University of Pittsburgh [3], Ferris et al. established 4 apps with 188 clinically validated skin lesions images. From images, 60 of them were melanomas. Three of four apps tested misclassified +30% of melanomas as benign. The fourth app was more accurate and it depended on dermatologist interpretation. These results raise questions about proper use of smartphones in diagnosis and treatment of the patients and how dermatologists can effectively involve with these tools.
In this study, we used a MATLAB (The MathWorks Inc., Natick, MA) based image processing algorithm that uses an RGB color dermoscopy image as an input and classifies malignant melanoma versus benign lesions based on prior training data using the AdaBoost classifier [5]. We compared the classifier accuracy when lesion boundaries are detected using supervised and unsupervised segmentation. We have found that improving the lesion boundary detection accuracy provides significant improvement on melanoma classification outcome in the patient data.
Outcomes and temporal trends among high-risk patients after lung transplantation in the United States
