December 14, 2018

Powerful Predictors: Diagnosing Cancer in India Using AI

Tubes to detect cancer

TL:DR;

  • Dr. AI reporting for duty. Mission? Provide an alternate solution to the shortage of trained specialists.

In India, approximately 74,000 women die from cervical cancer every year, accounting for 1/3rd of the burden of cervical cancer deaths globally. Less than 10,000 pathologists must service a population of 1.3 billion people. Efforts to increase the number of trained specialists has had limited success, leaving these cancers largely undiagnosed.

You might also like...

Aindra Systems, a tech start-up from India, has developed an innovative workaround. Leveraging artificial intelligence (AI) to provide point-of-care cervical cancer detection, Aindra is demonstrating how strategic use of automated technologies has the potential to democratize access to quality care in low and middle-income countries (LMICs).

When Adarsh Natarajan, founder of Aindra, participated in a workshop hosted by the Consortium for Affordable Medical Technologies (CAMTech), CAMTech Director Kristian Olson took notice. “Right from the get go he [Adarsh] had this vision about how AI technologies could be used to enhance accessibility of care to the over 5 billion people who don’t have access to quality healthcare,” recalled Olson, who has served as advisor to Aindra for the past 3 years. “Where Adarsh is leapfrogging is recognizing that there is a need and we are not going to overcome it with human resource training alone.”

Aindra’s end-to-end diagnostic platform was developed using human-centered design, an iterative approach to design to ensure a final product that meets the needs and circumstances of end-users. Biological samples are stained by an autostainer, converted into a digital image, then analyzed by AI algorithms to differentiate between cancerous cells and healthy cells. This three-part process is done onsite at a clinic, eliminating the need for the tedious process of manual staining and the transport of large batches of samples to distant laboratories. As a result, a patient is told if they have cancerous lesions within 1-2 hours instead of 5-6 weeks.

To Olson, it made a lot of sense that Adarsh and his team were focusing their platform on cervical cancer detection. “Cervical cancer is a really interesting target because early recognition of cancerous cells can result in effective treatment with low cost modalities,” said Olson. Additionally, the type of artificial intelligence they are deploying, deep learning neural networks, are particularly effective at image classification, with well-studied applications in dermatology and ophthalmology.

“Aindra is addressing a critical problem faced by many LMICs — a chronic shortage of human resources for health.”

Aindra is addressing a critical problem faced by many LMICs — a chronic shortage of human resources for health. “AI is creating diagnostic capabilities where human intelligence wasn’t capable before because of the volume and accuracy,” said Olson. The platform can sift through large numbers of samples and only escalate high-risk cases, complementing the work of overburdened pathologists. In a country with a high patient-to-pathologist ratio, this has the potential to increase workflow efficiency, reduce costs, and save lives.

Aindra’s platform makes a strong case for the use of AI to address global health challenges, but Olson sounds a note of caution. AI-assisted technologies require vast amounts of labeled, high-quality data to train the AI algorithm, data that doesn’t exist in many low-resource environments. In the case of healthcare, poor quality data could lead to fatal outcomes.

Adarsh realized they had to control the actual data inputs for their technology to work. “It would have been much easier if all they had been working on were the AI algorithms to identify cancerous vs. non-cancerous cells,” said Olson. “But really starting with how they systematically stain pap smears was an ‘ah-ha’ moment.”

As the use of AI surges, there is increasing awareness that AI solutions need to be built to solve a specific problem in a specific context. In the case of LMICs, issues including scalability, ethical and legal considerations, and efficacy will need to be addressed. However, where AI can address critical human resources for health needs, there could be immense impact. AI-powered cervical cancer detection, in combination with widespread HPV vaccination, has the potential to prevent deaths from cervical cancer globally — an historic achievement. Companies like Aindra could play a key role in reducing the burden of disease in populations with poor access to high quality care and set the stage for future entrepreneurs from LMICs to develop AI-assisted technologies using human-centered design.

People in this Article

Keep Reading

Next Up

Next Edition | January 7, 2019