AI Monday – Berlin – 28/10/2019

Healthcare is emerging as a prominent area for AI research and applications. Almost every area across the healthcare industry will be impacted by the technology. Image recognition, for example, is revolutionizing diagnostics. Pharma companies are experimenting with deep learning to design new drugs. Healthcare AI startups have raised investments, topping all other industries with AI deal activity.One of the biggest hurdles for artificial intelligence in healthcare will be overcoming the inertia to overhaul current processes that no longer work and experimenting with emerging technologies. AI faces both technical and feasibility challenges that are unique to the healthcare industry. Generating and opening up new sources of data promises to be revolutionary.These facts and trends led us to focus our AI Monday on Healthcare and AI. Berlin is a hotspot for AI and Healthcare with a tremendous number of startup companies, researchers and also industry located here.And the location was chosen with purpose to fit the topic: The Bernstein Center for Computational Neuroscience focuses on experimental, theoretical, and clinical research of brain function and dysfunction, and technical applications like brain computer interface, machine learning etc. Its internationally recognized for its strong expertise in computational neuroscience.With 6 speakers we have had a busy but exciting agenda.Anne Schwerk from the well known German Research Center for Artificial Intelligence (DFKI) opened up with exploring the potentials and challenges of AI driven medicine, specifically as its very data driven. Open innovation, open data & open science are the crucial enablers for data-driven medicine.Dr. Moritz Augustin is leading the AI team of Tiplu, a software company supporting hospitals. Stationary hospital stays in Germany are payed according to diagnosis related groups. These depend on diagnoses and procedures which must be coded correctly by the hospital to ensure its effort is appropriately compensated. Several companies offer software that optimizes medical coding using benchmark statistics – a questionable approach since this could lead to overpayment. Tiplu’s solution aims for right-coding and searches patient cases, particularly unstructured texts like medical reports, to identify the correct diagnosis and procedure codes. In his talk Moritz present a practical solution to this natural language classification task. Specifically, they train recurrent neural networks to recognize billing codes in unstructured medical reports in German language. They constrained the model complexity for applicability within the (old) computer technology standards in German hospitals.Dr. Michelle Livne from ai4medicine talked about approaches to leverage AI in identifying the right treatments for stroke patients – one of the leading causes of deaths worldwide with about 15 million strokes and 6 million deaths every year. Instead of a time-based statistically derived generalized paradigm approach, she and her team developed an individualized approach based on Neuroimaging.Dr. Andreas Lemke and his team from mediaire are trying to Disrupt Radiology. Algorithms are outperforming radiologist in specific tasks still almost no radiological practices are using those. This is due to not enough annotated data for training the algorithm. Mediaire circumvents these problems by using automated quality assurance algorithms and hybrid models incorporating Deep Learning architecture. They also have developed a way to learn from costumer annotations and use augmentation for training.Enes Hoşgör, Ph.D. shared his ideas and solutions under the umbrella of his company caresyntax, to revolutionize surgery. While in Sports we have understood and benefited from technology and use analytics to improve performance, we don’t do this much in hospitals. With surgical.ai he is digitalizing the surgery, leveraging Data to Reduce Surgical Risks plus adding AI to recognize mistakes or bad performance while attempting to overall improve quality.Finaly Sebastian Niehaus from AICURA medical GmbH shared his findings of a study they conducted on incorporating prior knowledge into deep learning workflows for CT image segmentation. He compared deep learning and expertise-based approaches, as well the performance of different models.The event was concluded with drinks and pizza.