Pune start-up that is fighting COVID, other diseases with AI-driven radiology service

Reading of a scan is a time-consuming activity that culminates with a radiologist going through 200 to 2,000 of images to come to a diagnosis. This is a cognitive process, and requires many years of experience, to reach a conclusion about a case. Add to this the fact that India has, according to several reports, around 10,000 radiologists catering to a population of 1.3 billion. Using technology, DeepTek is attempting to ease the stress by enabling experts to analyse and report faster. The platform allows the automation of repetitive tasks, enabling experts to focus on the disease, and creation of structured reports with just a few clicks, thereby minimising the efforts in writing or dictation. The platform reduces the margin of error that could be caused by visual fatigue or technical reasons.

“If, for instance, there is a CT study having COVID-pneumonia like pattern, AI has the ability to segregate the study with classic imaging findings from normal scans. It also enables experts to quantify lung lesions. Critical studies are moved up the list so that reports can be delivered earlier for these, this function of triage is critical to ensuring quick reports to critical patients. Triaging is usually done in a war setting — if there is a wounded soldier and there is another who is critically wounded, which is the one which needs to move first?” says Dr Kharat. “For the radiologist, their imaging worklist is actually a virtual war room where they are fighting against time to deliver critical reports to ensure quick delivery of care. Here, in the same way, the AI does the screening and segregates the critical studies and allows the experts to act quickly on them,” he adds.

DeepTek, which was founded less than three years ago, has generated a revenue of $1 million in its first year of going live. It handles more than 2,000 scans every day and is used by more than 150 hospitals and imaging centres across India. “Onboarding a center to provide this service is done remotely and these imaging clinics can be onboarded within 15 minutes. The team is enabling subspeciality reports, critical studies triage and virtual radiologist consultation in almost all corners of the country making imaging accessible and available. We are adding four to five hospitals and imaging centres per week,” says Patil.

Dr Kharat, who has 18 years of experience in practicing radiology before co-founding DeepTek, adds that the present Covid scenario has resulted in, apart from RT-PCR, High Resolution CT (HRCT) being recommended to assess the lung condition, score the lung involvement and exclude other complications or preexisting disease. “That has suddenly stressed medical imaging clinics and hospitals. Initially, the demand seemed only in Maharashtra state, however, it seems to be a pan-India as we get enquiries from all across the country. Enabling the imaging experts with the power of the platform to ease out the pressure is the best we can do in these trying circumstances” says Dr Kharat.

DeepTek is also working in the screening of tuberculosis from digital chest X-rays — which kills 1.4 million people annually. For TB, an X-ray chest is an important component of screening patients. Implementing AI for screening in such situations is helpful as AI currently works well as a point solution. The solution has also been audited by organizations supported by WHO such as Stop TB Organization and Friends of International Tuberculosis Relief (FIT).

The company has two major investors from Japan and a number of other smaller strategic investors. Having raised their seed round, they are planning to look for Series A funding in six to eight months “for strengthening the platform and developing AI models across modality, and varied disease spectrum”.

“This industry is at an early stage; it is an innovation in process. We have 7 patents, each defining a unique process which allows experts to create value. There are several challenges to be addressed while creating AI models. If you develop AI models for one hospital data, there is a chance that it may not work for another hospital dataset initially unless customized. Models need to be robust and they need to be integrated well in the existing workflows to ensure adoption. It will be a journey to cover the space and create comprehensive value,” says Patil.