Skin tumor diagnostics

The problem

  • More people are diagnosed with skin cancer each year in the U.S. than all other cancers combined. Skin cancer is the fourth most common cancer worldwide.(1)
  • Melanoma is the most aggressive skin cancer. For patients with early stage and localized melanoma the survival rate is 99%; yet when melanoma has metastasized it is more resistant to current therapies and the survival rate drops to 15-20%. Progression from early to metastatic stages can occur in months, making the prompt diagnosis of melanoma particularly important.(2)
  • Tissue biopsy is the only definite diagnostic test for skin cancer. With the increasing demand for quality healthcare globally and increased health screening, the number of pathology samples to be analyzed is also increasing. Thus, pathology experts are required to review non-cancerous specimens in more than 90% of cases.
  • The number of practicing pathologists retiring coupled with the years of practice required to achieve an expert level of competency is causing a drastically increasing shortage of expert pathologists worldwide.(8) Delayed responses and the high cost of pathologist opinions result in reduced time to implement proper care at the early stages of disease when the tumor is likely to respond to treatment.
  • Solutions helping to instantaneously differentiate benign neoplasms from potentially cancerous lesions can radically increase throughput, shorten response times and reduce costs.
  • Access to expert pathology opinions is limited in low-income countries, with less than 30% having access to pathology services in the public sector as compared to more than 90% in high-income countries.(1) Applying digital pathology and AI algorithms to cancer diagnostics is the only viable option to improve access.
  • According to studies in the U.S., pathologists provided incorrect diagnosis in 25% of cases on average, and expert pathologists sometimes disagreed with each other in up to 75% cases, particularly when the case is not clear-cut; and in up to 65% cases they disagreed with themselves. And general pathologists are commonly even less accurate than expert pathologists, which generates a higher demand for second opinions and expert consultations worldwide.(4)
  • Diagnostic errors contribute to approximately 10% of patient deaths and are the leading cause of medical malpractice payouts.(5)
  • Skin cancer is the most common cancer type in the U.S. with more than 500 thousand cases estimated in 2018 (including non-melanoma skin cancer), which is two-times higher than breast cancer, the second most common cancer. One in five Americans will develop skin cancer by the age of 70 and represents the fourth most common cancer type worldwide, with an incidence of 1.3 million. The incidence of various skin cancer types grows at 2-9% annually. (1),(6)
  • For very early-stage melanoma limited to the site of origin, the 5-year survival rate is 99%. Survival for melanoma invading nearby lymph nodes is 63%. If melanoma has metastasized, survival rate is 15-20%.(2)
  • 5% of skin tissue biopsies in the U.S. are diagnosed as melanoma.(3)
  • According to a study in 10 U.S. states conducted by Elmore, the discordance rates between pathologists assessing skin biopsies ranged from 57% to 75% in middle-ground classes, including disagreements among themselves ranging from 37% to 65%.(4)
  • The number of active pathologists decreased by 11% from 2010 to 2015, 63% of active pathologists are age 55 or older, and 44% of all pathologists work overtime.(8)
  • The annual cost of treating skin cancers in the U.S. is estimated at $8.1 billion in 2000s, with approximately $4.8 billion for non-melanoma skin cancers and $3.3 billion for melanoma.(7)

(1) International Agency for Research on Cancer, WHO

(2) American Society of Clinical Oncology

(3) BMJ 2005 Sep 3; 331(7515): 481.

(4) BMJ 2017;357:j2813

(5) National Academies of Sciences, Engineering, and Medicine. Improving Diagnosis in Health Care. The National Academies Press, 2015

(6) Cancer Facts & Figures 2018, American Cancer Society

(7) Guy GP. Prevalence and costs of skin cancer treatment in the U.S., 2002-2006 and 2007-2011. Am J Prev Med 2014; 104(4):e69-e74

(8) 2016 Physician Specialty Data Report, AAMC

Our solution

Skin tumor diagnostics

AI-based automatic analysis of WSI

Our approach utilizes advances in computer vision and deep learning coupled with our insight of the distinctive properties of a histopathology image. A specialized neural network learns from example whole slide histopathology images (WSI) annotated by top dermatopathology experts. The trained model is then used to create an accurate, robust and fast histopathology diagnosis or second opinion.

Hardware agnostic

Our unique solution architecture integrates well with WSI from all modern scanners, despite significant color variations of images acquired from various equipment.

High accuracy on the wide number of diagnoses

Our network is capable of recognizing 49 different skin tumor diagnoses from over 2000 WSI annotated by top dermatopathology experts. Testing on a set of 200 standalone sets of skin WSI images created on a number of different scanning hardware demonstrated accuracy of over 96%; and on a certain sample population we achieved 100% accuracy, which is significantly better than the average accuracy of pathologists.

Cloud and local

Our solution can be deployed locally at the laboratory and remotely as a secure cloud service.

Histopathology images are created by using specialized chemical staining which results in different color distribution when compared to natural images. We transform WSI and normalized image colors to eliminate the distortion from staining quality variations.

WSI are extremely large, with a typical image size of 100,000 x 100,000 pixels. We use a hierarchical multi-staged classification approach by first applying coarse and then precise classification techniques to suspicious areas, making our solution one of the fastest on the market.

A decision on the existence of tumor in a specific location is highly influenced by its surroundings (context). Our proprietary Convolutional Neural Network architecture factors the increased field of view (context) into the decision about the local area

Results to date

Performance metrics

Accuracy

Annotated skin tumor images

Minutes per slide

Number of recognized nosologies

Benefits

How our solution can help you

Healthcare providers

Immediate second opinions for tissue analysis anywhere in the world, virtual pathology services in remote sites and access to consultation services. Pathology laboratory outsourcing and cost reduction.

Diagnostic centers and pathology laboratories

Fast and precise detection of cancerous tissue, cell parameters assessment and the probability of each tumor type. Automatic separation of slides containing cancer from those that do not. Enhanced productivity and reduced laboratory expenses. The possibility to process requests from other locations requiring pathologist opinions.

Digital pathology software providers

WSI analysis and interpretation solution can be integrated into Laboratory Information Systems (LIS) to increase workflow efficiency and improve diagnostics decisions and speed

Digital pathology scanning solutions providers

WSI processing software can be embedded into digital pathology scanning, storage and display solutions. Fast and quantitative evaluation of images, streamlined WSI analysis workflow with precise detection of the tumor-affected tissue for digital pathology.

Insurance providers

Our solution helps reduce or eliminate the potential for misdiagnosis of biopsies and avoids malpractice payouts, reduces the cost of skin cancer diagnostics, response times and subsequent healthcare expenses, and improves treatment decisions and patient care.

Research and educational organizations

Our cloud solution can be used in clinical research, allowing remote pathology review and undergraduate teaching of histology, residency/trainee pathology training and professional training in specialist areas of pathology.

Our team

Leadership

Alexander Chervony

CEO

20 years of top management experience, 18 years in investments. Founder of Group Delta Investments (industrial investments, sold in 2005). LL.M. from Moscow State University

Leah Levine

Managing Director U.S.

Over 15 years of corporate and start-up innovations and business development within strategy consulting, banking and multiple tech entrepreneurial roles. MS, Applied Mathematics and Economics from Moscow State University

Larisa Chervony

MD PhD

Prof. Dermatologic Oncology and Dermatopatology at Moscow State Academy of Postgraduate Studies. Leading expert in clinical and histological differential diagnostics of skin tumors with over 40 years of experience. Author of a scientific discovery(1), 5 patents, 2 inventions(2), 234 publications, 21 academic papers, 3 monographs(3). Co-author of WHO classification of melanocytic skin tumors

Simon Polak

CTO

14 years of experience in R&D in Machine Learning, Computer Vision and Natural Language Processing both in academia and industry (Intel, JustVisual, Viisights). MSc in CS from Technion and 5 years in PhD program in Hebrew University.

Mark Luchter

COO

25 years of experience as a technology and operational leader. Founder of L-protector Consulting, CTO and co-founder of CIS Networking, Head of global information security and network infrastructure at Teva Pharmaceuticals. BS in CS and Applied Mathematics from St. Petersburg State Marine Technical University.

(1) “Consistent patterns in DNA accumulation in cytoblast during dysplasia and malignant growth”

(2) “Early diagnostics of melanoma”

(3) “Melanocytic skin tumors”, 224p, ISBN 978-5-9704-3673-8; “Clinical and morphological diagnostics of skin diseases (classification)”, 432p, ISBN S-225-04104-3

From the blog

Check our latest news

Mechanomind participated in Camelyon17 Challenge

Mechanomind participated in Camelyon17 Challenge

During the summer 2017, Mechanomind participated in Camelyon17 - a challenge dedicated to evaluating AI histopathology image analysis for the recognition…

Read more »

We are partnering with digital pathology laboratories and clinical research organizations for algorithm evaluation, accuracy improvement and tumors type expansion

Let’s get in touch

Call Us

+1(347)989-4091

Email Address

y@mechanomind.com