MIT’s CSAIL Builds New Smart AI System for Assisting Doctors

We all have witnessed how beneficial Artificial Intelligence systems have been for various industries. There are several manual tasks that have been automated by leveraging artificial intelligence. There are however still some critical areas where relying on AI Systems is not a wise decision. One of them is the health sector where an individual’s life is on stake and we have seen that AI is not error-free. Due to these reasons, we haven’t seen a full-fledged AI in health operations.

MIT’s Computer Science and Artificial Intelligence Lab (CSAIL) have come up with a solution that may change the present landscape. This world-renowned innovative lab has built a machine learning system, through this the system can perform predictions on the task provided to it or it can decide to defer the decision making power to an expert of the field.

MIT CSAIL
MIT CSAIL

The unique feature of this system is that it learns when it should take the decision on its own and when it should defer to a human expert, for making this decision the system considers various factors such as availability of an expert, years of experience, and other relevant factors. This system is a perfect example of the intermingling of Artificial Intelligence with human beings to perform complex tasks.

MIT CSAIL-2
Source

This machine learning system consists of two main parts namely, a classifier that helps in performing prediction on specific tasks, and the other part is a rejector that performs the decision making of whether the task can be handled by the system’s classifier or it should ask for help from a human expert.

The AI system is capable of recognizing patterns from chest X-rays, these patterns help in finding different types of conditions pertaining to atelectasis which is a lung collapse and cardiomegaly where enlargement of heart takes place. In the latter case, the results of this hybrid system have bettered the performance by 8 percent compared to either of the Human or AI on their own performance.

With the introduction of this system, the aim is to reduce the load on the human experts for trivial tasks that can be taken care of by such intelligent systems. With this, the development team has claimed that the hybrid model has been able to reach a higher level of accuracy than the actual expectations. Surprisingly these results have been possible with much lesser computational cost and a smaller amount of training data.

Ad
Deep Learning Specialization on Coursera

This type of model is not just to the healthcare sector but it can be expanded to text/image classification tasks for content moderation on social medial posts and on various websites that deal with a huge amount of content on daily basis. A human moderator with this intellectual system will surely prove to be better than previous implementations.

The future of such systems have huge potential, currently tested with synthetic experts (Better tweaking of parameters can be done), the next step involves testing with human experts and also train the model for identifying different types of diseases through datasets of X-Rays and images. 

LEAVE A REPLY

Please enter your comment!
Please enter your name here