Artificial Intelligence has experienced a veritable boom in the past few years, promising to transform the world as we know it. Modern AI is not only able to recognize cats and people in pictures, it was also able to beat the best human player of GO, a game that has been notoriously difficult to master for machines.
Surprisingly, mainstream AI has not found its way into medical analysis yet. The two main reasons are that Deep Learning Networks, the engine of modern AI, require millions of samples or images to be trained properly - which is simply not feasible for clinical studies. Second, each study requires an optimal setup of the network, as information from one data type does not generalize to another.
At Denapsis AI, we have taken a different approach. Our research is rooted in the analysis of medical and biological data, with all its peculiarities, limitations, biases, and uncertainties. We thus taught AI how to learn all by itself and without supervision. Think about a child acquiring language - he or she will first pick up different sounds and combinations, and then interpret them. All unsupervised. And without millions of hours of speech to listen to. Likewise, our technology first finds the patterns in the data, localizes them, and then links it to any meta-information that is available, e.g. disease or disease status, phenotype etc.
And since every data set and disease is different, we customize our software for each data set and problem.