Recently, DeepTek.ai, a Pune-based medical imaging AI startup unveiled Augmento X-Ray, an AI solution for chest X-rays. This FDA-cleared AI-powered tool aims to alleviate radiologists’ workload by up to 50% and enhance chest X-ray reporting quality. 

It achieves this by using deep learning algorithms to detect abnormalities in chest X-rays—identifying, categorising, and highlighting suspicious areas automatically, aiding clinicians in making precise interpretations. 

Given the vast volume of chest X-rays annually reaching 1.5 billion, the tool responds to the critical need for accurate and timely interpretations. It addresses the dearth of expertise in the country, with a population of 1.4 billion people and only 20,000 radiologists.

In an exclusive conversation with AIM, Ashutosh Pathak, chief technology officer of DeepTek, provided significant insights about the company’s cutting-edge AI models in radiology, global expansion plans, recent FDA clearance, and strategic partnerships.

“We are developing AI models for various radiological scans like chest X-rays, MRI of the spine, and brain scans etc. Our AI deployment platform, Augmento also has features which enable it to integrate with deep tech models and other third-party AI models,” said Pathak. 

The company boasts a team of 200, with a dedicated tech team and a panel of 70-80 radiologists. DeepTek’s co-founder, Dr Amit Kharat is a radiologist with over 20 years of experience, bringing substantial domain expertise to the team.

DeepTek also boasts of a solid international presence with over 500 customers including in APAC countries and Japan. The company recently received FDA clearance for two of its products, Augmento and a chest AI model, paving the way for expansion into the US market.

Efforts in India

The company’s Genki solution, a public health screening tool, has been extensively employed in India, particularly by the Chennai Municipal Corporation for over four years. “It is deployed in vans with portable X-ray machines and can swiftly detect tuberculosis or any chest abnormalities,” said Pathak, saying that this can be deployed in remote areas and without the immediate need for a radiologist’s presence. 

The solution significantly enhances patient detection capabilities, ensuring a larger population coverage. Pathak also elaborated on their expansion plans, mentioning their collaboration with the Tamil Nadu state government, aiming to cover multiple districts and eventually the entire state. 

Additionally, DeepTek is partnering with various state governments and city bodies in India, aligning with the country’s commitment to combat tuberculosis and eliminate it by 2025.

Pathak also highlighted the uniqueness of their CXR Analyzer in India, emphasising its organ-based approach, a departure from typical pathology-specific AI models. This particular model, the only US FDA-cleared one in India with an organised approach, covers a wide range of pathologies within chest X-rays, assisting radiologists in identifying and categorising abnormalities. Unlike models focused on one or two pathologies, this approach encompasses various abnormalities present in chest X-rays, benefiting radiologists by streamlining the identification process.

While the model aids in detecting, categorising, and localising abnormalities, it doesn’t replace the radiologist’s final diagnosis. Instead, it enhances their efficiency by helping them prioritise AI-flagged abnormalities, improving the turnaround time for critical cases.

Tech Stack & Partnerships 

Pathak mentioned that their’s is a slew of proprietary models developed using the unit architecture, primarily programmed in Python and trained on platforms such as PyTorch and TensorFlow. Additionally, they use React for the front end and Java with MySQL for the back end.

“Augmento however, is predominantly a Java-based enterprise application with a React front-end, Java middleware using Spring Boot, and a MySQL backend,” Pathak said.

Deeptek benefits from its collaboration with NVIDIA, Microsoft Azure, Google Cloud, and AWS as their models are GPU-powered and cloud-agnostic meeting various customer needs. 

He also reflected on the demand for large storage capacity because of the size of the files, saying, “We use bucket services, like Google Cloud Storage, AWS S3 and Azure Blob for storage because our application handles large DICOM files. Typically, an X-ray will be around 7 to 10 MB of file and CT MRI could be, 50 to 500 MB.”

The company derives its training dataset comprising millions of radiology scans sourced from their teleradiology services catering to 300 to 400 customers in India. “Over 3,000 scans have been used in this for the software validation… done by 27 plus US Board Certified radiologists,” said Pathak. This extensive validation process spanned over 4 months and yielded accuracy exceeding 95% for detecting abnormalities and categorising them within their AI model. 

What’s next? 

Pathak outlined DeepTek’s visionary strategy, emphasising two key product lines they seek to advance in the market. “One is the Augmento as an AI deployment platform. And second is the AI models that we are developing with the US FDA, which enables us to foray into the US market.”

Simultaneously, they maintain their concentration on the APAC region, citing their recent deployment in Singapore and their intention to broaden their impact across Southeast Asia. Pathak elaborated on their ongoing pilot initiatives in Singapore with hospitals like Changi General Hospital and Singapore General Hospital, stating, “Post the pilot, about a year in, we look into bringing other hospitals on the platform.”

DeepTek’s strategies include close collaborations with renowned entities like Shimadzu and BJC Healthcare, aiming to extend their services to leading hospitals in Thailand, Malaysia, the Philippines, and beyond. They are also intensifying efforts in the US market through partnerships with NTT Data and Shimadzu, aiming to provide Augmento AI and Augmento X-ray solutions to leading hospitals in the United States.