How AI, ML help screen breast cancer

Breast Cancer is the leading type of cancer in women. According to WHO, one in every 8 women in US is at the risk of developing a breast abnormality in her life time. It is well-established that early diagnosis is very critical in saving a life of a cancer patient. While most olden methods take time and requires physical inspection to detect the cancer, many get informed pretty late, which could cause trauma in their lives. Early detection of breast cancer is crucial and today’s technology can come to the rescue.

NIRAMAI says “AI has the potential to revolutionise healthcare delivery”. NIRAMAI Health Analytix is a Bangalore-based deep-tech startup addressing critical healthcare problems through automated solutions. At NIRAMAI, engineers have developed a new cancer screening software that uses machine intelligence over thermography images to enable a low cost, easy to use, portable solution and requires minimal human supervision. Their solution can detect cancer at a much earlier stage than traditional diagnostic methods and self-examination and can therefore improve survival rates. This method of breast cancer screening can detect tumors 5 times smaller than what clinical exam can detect, is non-contact, painless and free of any radiation, apart from being low-cost, and universally accessible. Using this solution, women of all age groups can undergo frequent screening without any side-effects.

Dr Geetha Manjunath, CEO & CTO, Niramai, tells us more about how this technology works, how it can be effective, how it can be detected earlier and lastly, how the non-invasive can help more women around the world.

DC: How does Niramai use Artificial Intelligence and Machine Learning to detect breast cancer?

GM: Niramai has developed a new patented technique called Thermalytix, a novel computer-aided diagnosis solution to automate detection of early stage breast malignancies. The core of Niramai software consists of novel computer algorithms based on Artificial Intelligence and Machine Learning that analyse thermal images and generate quantitative interpretation. Use of Thermography for detecting breast abnormality was well known to doctors since decades. However, thermal images are very hard to interpret as there are 400,000 pixels per person and each of those pixels can be any of the 1000+ colors. So, just manual/visual interpretation to find early malignancies have earlier resulted in lots of false positives. We bring in AI to this arena. Using computer vision techniques as well as raw temperature values, we are able to get accurate analysis of these thermal images.

Niramai solution uses off the shelf thermal cameras and in-house developed Thermalytix software to generate a thermal analysis report automatically, that a radiologist can review before making a final observation on the patient. Our ML algorithms now enable a low cost, easy to use, portable solution for detecting breast cancer which is effective in all ages of women and does not require expert skills to operate the system. The solution is also non-contact, non-invasive and privacy aware.

DC: How the technology has evolved and how it benefits now and in future?

GM: Breast cancer is the leading cause of cancer deaths in India, particularly among the urban population. Statistics show that the mean age of Indian women who get affected with breast cancer is much lower compared to western countries. Mammogram, the most common modality used for the diagnosis of the disease, has low sensitivity in women under 45 years of age and those with dense breasts. While ultrasound scans are good, they are largely subjective, and the results vary from radiologist to radiologist. Niramai test works on women of all age groups, and does not involve any radiation unlike mammography.

In order to create this technology, we collaborated with multiple reputed hospitals and expert radiologists and created a database of thermal images and associated ground truth using other standard modalities. We have also used several novel features for training our machine learning models which have resulted in multiple US patents.

DC: How receptive has been the market to AI in healthcare?

GM: There is increased acceptance of AI in healthcare systems today. AI has been used for improving operational efficiency of patient care delivery workflow in many large hospitals. Expert radiologists are starting to see AI based software as tools that can help them be more productive and more confident of their diagnosis since all the data related to the patient can be quickly analyzed prior to making a treatment decision. To quote the concluding remark at the largest radiologist conference in Chicago, RSNA 2018: “In the near future, Machines will not replace Radiologists, but a Radiologist using AI will replace Radiologists who don’t.