LThe foundation of artificial intelligence (AI) was laid half a century ago, and medicine has been an important field of experimentation from the very beginning. The first artificial neural network was proposed in 1958. Inspired by the function of the brain, this algorithm connects mathematical functions or “neurons” to each other. Each neuron assigns a value depending on the size of its multiple inputs.
When delivering data, the network adjusts the importance given to each neuron to deliver the solution. By multiplying the number of examples, the algorithm provides the correct answers to the new situation: it learns. This work has been in the laboratory phase for a long time, but recently things have changed, allowing access networks for large databases to be trained, with high-performance algorithms and enhanced computers repairing millions of neurons.
If AI classifies images of objects or animals more reliably, it is possible to detect lesions in medical images. Based on the current results, AI’s ability to detect subtle pathological phenomena (classification of skin tumors) or their evolution (modification of lesion size), without fatigue, will reach or exceed human performance in quality and speed. Decline. This opens up new therapeutic perspectives.
“AI’s proven effective, but tools are rarely used in the medical environment”
Automatic measurement of brain lesions that appear between the two consultations allows the neurologist to customize the treatment of his patient. Extract and analyze the records of thousands of cranial trauma patients admitted to CHU Hospital over the past five years, classify them, analyze relevant images, extract signatures predicting the severity of the attack, and establish optimal treatment protocols. The potential of AI has been proven, but tools are rarely used in the medical environment.
What are the barriers to the widespread deployment of AI in health, and where are our research efforts? AI models are faster and more reliable than the human operator, and the examples used in their training for all patients can only be generalized if they represent a high quality, multifaceted and versatile event. Our efforts focus on creating large, compatible databases that take into account acquisition conditions and the diversity of patient recruitment pools.
You should read 60.4% of this article. The following are for subscribers only.