Artificial Intelligence Versus Clinicians in Disease Diagnosis: Systematic Review (Preprint)

Abstract

Artificial intelligence (AI) has been widely used in medical field, facilitating the development of medical services. This passage illustrates basic principles of AI application in medicine, summarizes some outstanding achievements of such application and discusses its merits and drawbacks. The passage is also an attempt to look into the future of AI in medical field. Keywords Artificial intelligence; disease diagnosis; treatment; deep-learning Artificial Intelligence in Disease Diagnosis and Medical TreatmentArtificial Intelligence (AI) is a field of program-based computer science that can simulate human's mental processes and intelligent activity and enable machines to solve problems with knowledge. In the information age, AI is widely used in the medical field and can promote medical development. AI may optimize the care trajectory of chronic disease patients, suggest precision therapies for complex illnesses, reduce medical errors, and improve subject enrollment into clinical trials. [1] There are currently two common methods in AI. An expert system is a computer system that generates predictions under supervision and can outperform human experts in terms of decision-making. It consists of two interdependent subsystems: a knowledge base and an inference engine. While the knowledge base contains accumulated experience, the inference engine (a reasoning system) can access the current state of the knowledge base and supplement it with new knowledge. Expert systems can create critical information for the system more explicit, make maintenance easy and increase the speed of prototyping [2]. However, expert systems are limited in terms of knowledge acquisition and performance.The second method is machine learning. This is the core of AI and is a fundamental approach to making computers intelligent. Machine learning requires huge amounts of data for training. This improves their performance step by step during the process. One of the focuses underlying machine learning is parameter screening. Too many parameters can lead to inaccurate entries and calculations. Therefore, reducing the number of parameters can improve the efficiency of AI, but it may also lower its accuracy. However, one of the key objectives of AI is to outperform people via selfstudy in challenging fields without any prior knowledge.Google's DeepMind was recently updated with new version of AlphaGo Zero, which started from scratch by relying completely on itself for intensive study and selftraining without learning from any chess manual. It then defeated AlphaGo that had studied a large number of chess manuals before by 100 to 0. The study ushered in a new era of AI that it is free from human cognition and can discover new knowledge and strategies itself [3]. In China, a robot designed by iFlytek. passed the 2017 Licensing Examination of Medical Practitioners. As a result, it will assist doctors in medical diagnosis and treatment.In machine learning, neural network is an algorithm. The network acquires knowledge through many examples in med...