AIndra, a health-tech AI startup dedicated to the mission of building innovative products and technologies in computational pathology was established in the late 2012 to build state of the art medical devices for screening cervical carcinoma, and since then it has been catering to other critical illnesses, all in the “AI way”. As the startup proudly claims, they create products in the area of artificial intelligence and computer vision.
With a smart team of 10 including scientists, researchers and engineers, they have a string startup hustler culture and are extremely driven towards their vision of democratising access to quality healthcare. “It was this vision that drove us to use deep technology as an enabler to create products that can create a huge impact on the society”, said Adarsh Natarajan, CEO and Founder, AIndra Systems.
AIndra Systems’ data driven clinical pathology
Based out of Bengaluru, India’s Silicon Valley, the startup has been fascinated with the world of artificial intelligence, and has been developing interesting products that solve big problems. With a keen ambition, the team is looking forward to have a platform for data driven approaches in pathology. “Our aim is to have devices that can digitise the information in slides and our proprietary algorithms can then use this information for clinical inference. This is all connected over the backbone of internet to get confirmation from a pathologist for a clinical sign off”, said Natarajan.
Core technologies like AI and analytics powering ‘CervAstra’
Explaining about the technologies used at AIndra Systems, Natarajan said that the company deals with lot of unstructured data, mostly images. And as it has been proven in last few years, deep learning in particular is very effective at handling this. “Consequently, we too use deep learning to build our core technology of clinical inference from images”, he said. “Apart from this, we use the latest advances in optics, electronics and mechanical systems, and have built a full stack solution comprised of hardware driven by the AI platform”, he added.
He further explained that their AI powered cervical detection system, CervAstra, is a powerful system that reduces cancer screening time considerably. “A rural woman today will take anywhere between 5-6 weeks to receive a report after her screening sample is collected, which can be attributed to work overload and low number of pathologists. CervAstra provides the report at the point of care, thus aiding faster screening and affordability”, he said.
Once the pap sample is collected, it is processed through the first device of CervAstra, which is an autostainer called Intellistain. The stained sample is then scanned using VisionX. This image can either be used for Telepathology or is passed through the third system, Astra, which is a powerful algorithm that classifies the sample into “Normal” or “Cancerous”. A report is generated and then shared with the woman at the point of care.
Giving us a detailed pointer on each of these, Natarajan said that their AI based computational pathology platform comprises of three components.
- The first one being, Intellistain, the autostainer that stains the biological samples.
- The second component is the VisionX, which is a Whole Slide Imager that converts the stained slide into a digital image.
- The third component is Astra, the AI algorithms based on Deep Neural Networks. This algorithm is trained and built to distinguish cancerous cells from normal ones.
“All these three components can also be used separately as well as an end to end platform. We are building a platform that can be extended for building similar devices for other cancers as well”, he added.
Natarajan believes that the biggest challenge in rural India is access to an affordable screening, and it may become a complicated process owing to the below issues:
- Slides have to be transported from the area of sample collection to the area of diagnosis.
- Staining of slides is difficult and strenuous if the manual staining technique is used.
- The samples are brought in big batches to the diagnostic centres which increases the turnaround time for reports.
CervAstra therefore makes Cervical cancer screening affordable, reliable and accessible.
Growth story at AIndra so far
Out of the three parts required for CervAstra, prototypes of Intellistain have been developed while that of VisionX is in the final design phase and the algorithms are ready and undergoing fine tuning.
“The system is under the clinical study phase and is being validated at a tertiary government oncology hospital (Kidwai Memorial Institute of Oncology) and a private sector lab (RV Metropolis), which will be a good mix of rural and urban samples. We intend to test about 2000 women using the traditional method and AIndra’s method to establish the benefits of using CervAstra”, he said.
“We have filed a full patent application for CervAstra in India and have two PCT’s that are being converted to National Phase Applications in a few countries and markets around the world. We are currently undergoing ISO 13485 certification along with the CE certifications for the products”, he added.
The team aims to make computational pathology the go-to modality for screening of fatal diseases. “We are working on creating a suite of products that leverages the Astra platform to enable Point of Care diagnostics”, he shared.
Challenges of being AI startup
Natarajan believes that in many ways working on AI is simpler than that of many other fields. “Being primarily a software system, AI skips over a number of challenges that are faced by hardware startups. However unlike traditional software, AI strives to give results that are similar to that of doctors. And there are subtle differences between the field of medicine and computer science that will manifest in evaluation and interpretation of results”, he said.
He further added that consequently the published results in the field of AI may not translate to the results in field. “In a nascent field like AI, working without help of prior literature is quite hard and can be challenging to deal with at times,” he said on a concluding note.
Try deep learning using MATLAB