5.3. AI Doctor

With the advent of the big data era and the continuous improvement in computing ability, AI solutions led by DL-deep learning have made great progress in the medical industry. As an important branch of artificial intelligence, DL-deep learning is receiving more and more concern and recognition from both the scholarship industry and industrial community. DL-deep learning is a method of modeling data by using a deep neural network. The network automatically learns the hidden features of data layer by layer, and then carries out the corresponding tasks of classification, regression and segmentation. As shown in the figure below:

Model Diagram of DL-deep Learning

Behind the DL-deep learning lays a deep neural network, which, through simulating human brain neurons, can learn potential features from the source data, thus activating the neurons corresponding to the hidden layer and finally mapping the output. As shown in the above figure, convolution neural network, as the classic of DL-deep learning. takes partial human visual feelings as a basis and extracts features layer by layer by visual neuron nodes. Finally, partial features are fused at high level and mapped to the final output. The circulation neural network is good at modeling time series data and can do a better job on modeling patients' cases consisting of the time axis, so as to give a comprehensive judgment on the patient's condition by better using their cases and medical history.

DAPP in AI MEDICAL adopts DNN-deep neural network technology as its core, and the enhancement of the DAPP capability requires training by a large amount of data. AI MEDICAL provides an online training platform, and the data in the platform is open after users' authorization. It supports a series of DL-deep learning frames such as the mainstream Tensor Flow and Caffe, and the AI model trainer can only get the conclusion of the AI model without taking data. AI model training needs to hold a certain amount of AIM tokens, then will consume the tokens, and the consumed AIM tokens will be rebated to the data provider.

The AI model is the nerve system of the AI MEDICAL network. AI MEDICAL puts the AI model provided by the DAPP provider on the AI model chain, and the model has the ability to operate independently, and lets the model have incremental learning ability. Compared with the traditional batch machine learning algorithm, distributed online incremental learning is the general method that is more in line with reality. In general terms, distributed online incremental learning can continuously learn new knowledge from the newly added data around, and it overcomes the time and computing ability-consuming disadvantages in past data redundancy learning, which is more in line with the human development process. Generated by the combination of DL-deep learning and blockchain, distributed online incremental learning is to share data on the chain by using the data stealthies of the blockchain. With the increase of blockchain nodes, data is also increasing. At the same time, the existing model can be rapidly updated by using the online incremental learning. which means when the disease prediction is given, the model is also updating and growing. In conclusion, the distributed online incremental learning system can utilize the advantage of fairly recording events on the time line by using the blockchain, and overcome the disadvantages of traditional batch learning, which means it can adapt continuous learning and improving at the same time of giving prediction results for the patient.