Diagnosing Skin Cancer with Melody:
Deep Learning Telemedicine identifies skin malignancy
in a first to publish prospective Clinical Study
Boston. February 14, 2019
Bostel Technologies, LLC is a clinical stage healthcare company which is developing a unique technology for Skin Cancer diagnosis by sonification, utilizing image conversion to sound. Proprietary techniques allow all physicians to diagnose Skin Cancer on site by Telemedicine as an Artificial Intelligence (AI) powered Clinical Decision Support Device
Skin cancer, the most common cancer in US affecting 5 million patients annually. The most deadly form, melanoma, is expected to be diagnosed in 96,000 people, and will lead to as many as 7,000 deaths annually in the US alone. Early detection of melanoma is a clinical window of opportunity which allows an almost 100% survival. As of today, clinical accuracy in detection of Skin Cancer by all physicians is of a limited diagnostic accuracy due to the complexity of visual inputs embedded in a dermoscopy image, and its dependency on physician skills and subjectivity.
As an example, a 23 year old female approached a dermatologist (2018) with a small new mole on her thigh. It was diagnosed as benign. She then consulted a second dermatologist who used a computer vision system in order to categorize it as benign. A third opinion was sought by the patient and a trained dermatologist used dermoscopy in order to identify a malignant melanoma which was biopsied and diagnosed by histopathology as malignant melanoma. "This circumstance is not exceptional and represents the difficulty in diagnosing melanoma at its start, during the gold period of excision. In objective tests dermatologists achieve a mean sensitivity of 40% for MM detection , due to the complexity of visual inputs embedded in a dermoscopy image " says Dr. A Dascalu, dermatologist and Chief Scientific Officer of Bostel Technologies, LLC.
The Bostel technology uses a regular dermoscopic image which is transmitted and processed by Cloud Computing through a convolutional neural network, a class of deep learning. Image is further deciphered by a sonification technique, which amplifies detection accuracy of Skin Cancer. A recommendation to excise or do not excise is conveyed to the physician within seconds achieving a accuracy significantly better than clinician only diagnosis. The technology was developed by a team approach of experts in sonification, computer vision and clinicians.
A clinical study demonstrating the Bostel technology was recently published in Lancet EBioMed (doi.org/10.1016/j.ebiom.2019.01.028). This study demonstrates the successful in clinic field test and validated the Sonification technology. 63 patients were prospectively identified and biopsied, 28 of identified as skin cancer malignancy. The Sonification met its primary aims of Sensitivity (86%) and Specificity (69%). By adding an additional layer of vision analysis through an additional deep learning system both negative and positive predictive values were 89%.
"Bostel combined multi layered diagnostic platform is expected to accurately assist clinical decisions by increasing the availability to accurately diagnose skin cancer, especially melanoma, by all physicians" comments Dr. Dascalu. "Most of the global R&D is directed at treatment and not prevention of melanoma. In an age of new biologic treatments, which strain healthcare budgets, the focus should be redirected to preventive medicine. Bostel’s technology will increase accuracy of diagnosis and decrease unnecessary skin excisions of benign lesions".
Bostel Technologies, LLC
SOURCE Bostel Technologies