TookThe foundations of artificial intelligence (AI) were laid more than half a century ago, and medicine was a major area of use from the very beginning. The first artificial neural network was proposed in 1958. Inspired by brain function, this algorithm links mathematical functions or “neurons” to each other. Each neuron assigns a value that depends on the level of its many inputs.
When presenting data, the network adjusts the importance assigned to each neuron to provide a solution. By multiplying the number of examples, the algorithm provides the correct answer for the new situations: it learns. This work has been at the laboratory level for a long time, but things have changed recently, with access to large databases where networks can be trained, high-performance algorithms to accommodate millions of neurons, and increased computers. Is.
If AI can classify images of objects or animals very reliably, it becomes possible to detect lesions in medical images. Based on current results, AI’s ability to recognize subtle pathological cases (classification of skin tumors) or their development (modification of lesion volume) will reach, or exceed, human performance in quality and speed without being related to fatigue. will be even higher. Downfall. This opens up new therapeutic approaches.
“AI’s potential is proven, but the tools are little used in medical environments”
Automatic quantification of brain lesions that appear between two consultations allows the neurologist to personalize their patient’s therapy. Extraction and then analyze thousands of records of cranial trauma patients hospitalized in a CHU over the past five years, classify them, analyze associated images, extract signatures predictive of attack severity, and establish customized treatment protocols. The potential of AI is proven, but the tools are rarely used in medical environments.
What are the barriers to the widespread deployment of AI in health, and where are our research efforts? AI models, which are faster and more reliable than a human operator, can only be generalized to all patients if the examples used for their training are of high quality, numerous and represent the diversity of cases encountered. We do. Our efforts are focused on creating large, cohesive databases taking into account acquisition status and the diversity of patient recruitment pools.
You have 60.41% left to read this article. The following is for customers only.
Prone to fits of apathy. Music specialist. Extreme food enthusiast. Amateur problem solver.