Young Viksit Kumar lost his mother to Ovarian Cancer, which was detected at a stage too late for any recovery through chemotherapy. Had his mother’s condition been detected earlier, it could’ve perhaps been a different story altogether. This jolt stirred Vikas to switch roads from mechanical engineering to being a medical student, to currently being a Senior Research fellow at Mayo Clinic in Rochester.

Never looking back since then, Kumar has been spearheading the use of GPU-powered deep-learning for more accurate and earlier cancer diagnosis than the ultrasound treatment, buying more time to save many cancer-affected lives.

Deep Learning in Cancer DiagnosisHis efforts are directed towards breast cancer which is more common than ovarian cancer and hence draws more funds. The primary objective is to effect earlier diagnoses in the developing countries.

Delving deep into Deep Learning

While at his work on using ultrasound images for diagnosing pre-term birth complications, he realised that the ultrasounds could be made to classify breast cancer, as they picked up different objects. Delving deeper, he concluded that deep learning was a good solution, following which he dived into the automaton mode to know everything that he could know about building and use of deep learning models.

There was a drive behind that learning: This was a tool that could really help. – Viksit Kumar

As in help in the timely detection of Breast cancer, especially in the developing countries where the expensive mammogram machines are confined to the metropolis.

The NVIDIA GPU Cloud cancer diagnosis via ultrasounds are economically feasible, and even in developed countries where regular mammograms of women after the age of 40 aren’t a rarity, ultrasounds could be the key to diagnosing pregnant or to-be pregnant women of cancer, as they can’t be treated through a mammogram’s X-rays.

Going up the curve

Kumar’s progress with the deep learning tools is a positive curve as the configuration of a system for deep learning only takes some hours now as opposed to the two or three days that took him to do the job earlier. The local processing is done using the TensorFlow deep learning framework container from NVIDIA GPU Cloud (NGC) on NVIDIA TITAN and GeForce GPUs. The heaviest lifting, requires the work shift to NVIDIA Tesla V100 GPUs on Amazon Web Services, using NGC’s containers, which are optimized for maximum performance on NVIDIA Volta, Pascal architecture GPUs on-premises and in the cloud.

Once we have the architecture developed and we want to iterate on the process, then we go to AWS,” Kumar said, highlighting his plan. As of now, Kumar’s team is working both on training and inference on the same GPUs. Adding to this, Kumar has said that he wants to do inference on an ultrasound machine in live mode.

More Progress up ahead!

Taking his plans to reality, Kumar intends on hitting live patient trials within the next year, applying the technique, inching towards his goal to use ultrasound images effectively in early detection of other kinds of cancers too, such as thyroid, ovarian etc.

Kumar’s pathbreaking vision in the field of medical science is to beat cancer right when it’s young, catching it early and nipping it in the bud, which is a commendable feat in itself, once achieved. However, patience is the key to all things, and patience is what Kumar asks of his patients regarding their approach to this newfound technology. “It needs to be a mature technology before it can be accepted as a clinical standard by radiologists and sonographers,” he said.

The rapid progress of Kumar and his team’s work suggests that the day this technology gets officially formalised by clinical standards can’t be too far, and we hope it arrives sooner than soon, for the sake of all the cancer patients!