In Brief:
- Artificial intelligence (AI), computer programs that simulate human intelligence, is being used to diagnose and treat human disease
- AI can screen for cancer and evaluate tumors to complement the work of doctors
- It is challenging to gather enough data to allow the software to make reliable diagnoses
We live in a world driven by algorithms, deep learning, and convolutional neural networks. Sound convoluted? It is! But these are the foundations of artificial intelligence (AI) – promising technology that has made many of our day-to-day activities possible, from easy navigation to mobile bank deposits and more. Now, AI is beginning to revolutionize medicine with groundbreaking software that helps physicians diagnose cancer and other diseases.
Meet Your New Doctor: AI
Artificial intelligence, often called machine learning, refers to complex software that simulates human intelligence to achieve a particular goal. By analyzing large amounts of data and detecting patterns within them, computers can imitate how people solve problems and even reason. What’s more, computers running AI programs are able to learn from their past experiences.
In contrast to typical software that uses reams of code to tell computers (‘machines’ in AI speak) exactly what to do, AI programming relies on algorithms that make it possible for them to achieve certain outcomes without human direction and intuition. As they are exposed to more data, the machines’ ability to classify it improves and they are able to detect subtle differences more easily. The more data machines are exposed to, the smarter they get, and the better they can perform diagnostic decision making. AI has already shown promise in diagnosing melanoma, a particularly deadly skin cancer.
The basic idea is simple. Many photographs of melanoma lesions (moles and tumors) are scanned and evaluated by the machine for such characteristics as size, color, location and border regularity. Photographs of benign (non-cancerous) lesions are also scanned in and evaluated. As more data is gathered, AI machines can eventually tell which lesions are cancerous, and which are not.
Doctor Hazel
Mike Borozdin and Peter Ma, California-based software engineers, were devastated when a mutual friend lost his life to cancer. They decided to join forces to construct innovative software: Doctor Hazel. An accessible and low-cost program to allow screening for skin cancer, Doctor Hazel allows physicians to upload a picture of a questionable mole, and within seconds the program will indicate whether there is a need for further testing.
As Borozdin and Ma point out, dermatologists are among the busiest doctors. Doctor Hazel can triage faster, helping to identify those who are in need of urgent care so they can be sent directly to a specialist. In such situations, timing can be a matter of life and death. With early detection, chances of survival for a patient with certain kinds of skin cancer can be up to 95 percent, but this number significantly decreases if the cancer spreads.
AI vs Doctors
Currently, pathologists examine images of patients’ lesions and use their judgment to characterize them and to identify the best treatment. It’s an activity that combines both art and science, one that’s dependent on a clinician’s skill and experience. Studies have shown that AI results match the accuracy of board-certified pathologists and can sometimes exceed it.
Professor of biomedical data at Stanford University, Dr. Daniel Rubin, MD, MS, has created an AI program to facilitate the classification of lung cancer lesions. Programs for this purpose are designed solely to detect and stage the disease. “It’s the areas where clinicians don’t do so well,” says Rubin. “Making predictions about clinical outcomes, or helping tailor treatments to patients – those are areas where there’s clearly a need [for artificial intelligence].”
Challenges
All three scientists identified the same challenges facing AI diagnosis: securing FDA approval and gathering enough data for deep learning. Ma says the FDA looks for two characteristics: transparency and certainty. Unfortunately, it is difficult to guarantee one hundred percent accuracy because AI, like humans, is prone to error. These obstacles could also mean that it may be years before AI diagnostics reach the market.
In addition, AI programs must be exposed to large amounts of data if they are to be of optimum benefit. However, several major cancer centers do not share their data, which has proven to be a major struggle for software developers. “No one institution has a large enough amount of data on a population to build robust models that will be generalizable to practices in the United States,” says Rubin.
The Future
Despite the challenges that remain, software engineers are optimistic. Artificial intelligence will undoubtedly transform medical diagnoses and enhance the quality and speed of many industrial processes as well. “The great thing about AI, and why it has been so great for medicine, is that it doesn’t forget. You don’t lose knowledge, but you can also share knowledge from all over the world,” says Borozdin.
CONTENT EXPERTS
Peter Ma is the co-founder of Doctor Hazel and graduated from N.J.I.T. with a B.S. in computer engineering. Mike Borozdin is also the co-founder of Doctor Hazel and VP of Engineering at the mortgage software company, Ethos Lending. The engineers created Doctor Hazel at one of the numerous hackathons that they attended. After a mutual friend’s death from cancer, they decided to apply their AI knowledge to create software to screen for lesions. Since then, both have been featured in several expositions (Techcrunch, Intel AI Academy, Mobi Health News) and have demonstrated the abilities of their software at several conferences.
Daniel Rubin is a professor at Stanford University who currently teaches biomedical informatics as well as biomedical data science. His main focus is imaging informatics which investigates how computers can help physicians analyze images. He is a graduate of Stanford University School of Medicine and a board-certified radiologist.
Works Cited
- “He’s Chasing Genius: Peter Ma ’07.” Reel Talk: History Professor Neil Maher on Oscar-Nominated ‘Hidden Figures’ | NJIT News, 16 Mar. 2018, news.njit.edu/hes-chasing-genius-peter-ma-07/.
- Admin. “AI Helps with Skin Cancer Screening.” Qualify for Free Software – Student | Intel® Software, Intel, 13 July 2018, software.intel.com/en-us/articles/ai-helps-with-skin-cancer-screening.
- Stanford University. “Artificial Intelligence Used to Identify Skin Cancer | Stanford News.” Stanford University, 3 May 2018, news.stanford.edu/2017/01/25/artificial-intelligence-used-identify-skin-cancer/.
- “How Artificial Intelligence Could Help Diagnose Cancer and Predict Survival.” Stanford Medicine, stanmed.stanford.edu/2017summer/artificial-intelligence-could-help-diagnose-cancer-predict-survival.html.
- Ma, Peter. “Doctor Hazel: A Real Time AI Device for Skin Cancer Detection.” Qualify for Free Software – Student | Intel® Software, Intel, 23 July 2018, software.intel.com/en-us/articles/doctor-hazel-a-real-time-ai-device-for-skin-cancer-detection.
- McCarthy, John. What Is AI? / Basic Questions, 1 Jan. 1970, jmc.stanford.edu/artificial-intelligence/what-is-ai/index.html.
Image Credits:
By Team Graphic Designer: Macafie Bobo
Chief Editor: Akila Saravanan
Creative Team Manager: Lucia Tian
Team Editor: Anika Prakash
Team Graphic Designer: Macafie Bobo
This article was written by Sanya Jain. As always, before leaving a response to this article please view our Rules of Conduct. Thanks! -cSw Editorial Staff