Not very long ago, India’s sex ratio numbers were alarming. It was 896 per 1,000 males in 2015-17, after falling from 898 per 1,000 males in 2014-16. A decade ago, the gap in ratio was significantly more, raising real concerns about possible female foeticide and consequently, gender determination tests being carried out in remote parts of the country despite being illegal.

Several NGOs have been rallying the cause of this widening gap, in addition to efforts by the government to spread awareness in society on the perils of infanticide and foeticide. Data collected over the years have further supported these claims. Government initiatives like Beti Bachao Beti Padhao have been actively propagated across the country – so far, 161 districts have been covered in two phases and the remaining 479 will be covered in the third phase.

Latest data on sex ratio at birth (SRB) has thrown some surprising results – the ratio has gone up – it now stands at 931 females to every 1,000 males in India’s rural pockets, contrary to common perception of urban India being more welcoming to a girl child.

While policy efforts continue in this domain, isolated incidents such as the recent report of no girl child being born in three months in Uttarakhand’s Uttarkashi district stoke the same old fears.

But, technology can play a critical role in mitigating the risks of danger to the unborn foetus and expectant mothers.

AI To Restrict Gender Determination

Bangalore-based Inferencia Logic, which builds intelligent machine vision systems, developed a method to mask gender imaging metrics during an ultrasound. Co-founder Paras Agarwal explained how the team relied heavily on AI technology to restrict gender determination of a foetus by doctors. These AI systems can mask specific regions in the ultrasound videos of pregnant women before it gets displayed on the monitors or being stored in drives. Special access to unmask these regions is provided with the help of passwords.

AI is mostly about data and when it comes to solving problems in the healthcare sector, acquiring data becomes a notable challenge. However, doctors at prestigious medical research institutes in Bangalore are helping Paras and his team (subject to all necessary permissions and compliances) in receiving a high volume of ultrasound videos and help them in understanding how to decode ultrasound videos. Using this data, the team is working on creating a large pool of labelled data for training their algorithms.

Accurate Yet Light Algorithms For Fine Tuned Results

The challenge pertaining to designing the system is two-fold, claims Paras. Firstly, the algorithm has to be highly accurate so that it masks the ultrasound with the foetus’ reproductive parts flawlessly. Secondly, algorithms should be extremely light so that it can be integrated directly in ultrasound machines to give desired frames-per-second (FPS), working real-time on the device without any internet or cloud dependencies. If desired FPS is not achieved, doctor will experience a lag in videos during a diagnostic ultrasound procedure. To achieve this, Inferencia Logic’s engineers used Google’s state-of-the-art research work called as ENAS to dump out optimal architecture of the model. Eventually, fine-tuning the resulting model manually and retraining on a larger dataset provided a higher accuracy-vs-speed ratio. Continuous infeed of labelled data and retraining the models will give desired accuracy before the final implementation.

“Currently, we are in talks with one of the topmost medical equipment manufacturer in India to integrate these models. We are looking forward to start working on clinical trials and creating a significant social impact in India by mitigating gender inequalities using technology,” said Paras.

Source: NASSCOM Community