Deepfakes have gained a lot of negative attention recently. Be it the hugely criticized DeepNude AI app which removes clothing from pictures of women or the FakeApp that swaps the faces of celebrities with porn stars in videos.
But the AI technology behind deepfakes and deepnudes isn’t entirely evil. Deep-learning algorithms are excellent at detecting matching patterns in images. This capability can be used to train neural nets to detect different types of cancer in a CT scan, identify diseases in MRIs, and spot abnormalities in an x-ray.
But it’s not that easy
While the idea of implementing deepfake AI for medical purposes sounds great, researchers don’t have enough data to train a model — simply because of privacy concerns.
At a time when privacy is a primary concern for the masses, no hospital or patient is willing to give away this precious data.
The second challenge is the requirement of deep-learning algorithms to train on high-resolution images for obtaining the best output.
But synthesizing such high-res images, that too in 3D, requires a lot of computational power. Conducting such experiments would require special and expensive hardware and it isn’t feasible for hospitals to implement it at a large scale.
GAN to the rescue
This is where the Generative adversarial network (GAN) steps in. GANs don’t require a lot of external data as they can synthesize more medical images that are similar to the real ones, thereby multiplying the data set required to train the AI models.
Moreover, GANs have the ability to produce highly realistic images that can also be used for various medical diagnosis.
Researchers from the Institute of Medical Informatics at the University of Lübeck have come up with a process that has fewer requirements.
It has two stages: First, the GAN generates the entire image in low-resolution. Then it generates the details of the image at the right resolution and works on one small section at a time.
Through these experiments, the researchers were able to generate realistic high-res 2D and 3D images using less computational resources. The best part is that they were also able to do so without increasing the expenditure (irrespective of image size).
The aftermath of apps using deepfake AI has caused several states to put up new laws against it, such as the recent one put by New Virginia where one could face up to a year in prison for using deepfake revenge porn.
But the recent development shows that technology can also be used for the benefit of humanity.