Before discussing futuristic applications of computer vision in healthcare, let us talk a little about how computer vision works. Although, the ability to make machines “see” a still image and read it, is related to human’s ability to see, the machines see everything differently. For example, when we see a picture of a car, we see car doors and windows and glasses, color, tires, and background, but what a machines sees is just a series of numbers, that simply describes the technical aspects of the image. Which does not proves that it is a car.
Now, to filter out everything and to arrive to a conclusion that it is a car, is what Neural Networks do. Various Neural Networks and Advanced Machine Learning Models are being developed and tested over the period, massive amount of training data as being fed and machines, now have achieved a level of accuracy.
How AI could benefit Health Care Industry:
There have been discussions on how AI could help various industries and Health Care is one of the most talked. There are many ways AI could support the industry.
- AI could engage in repetitive tasks, like examining vast amount of X-Rays, Data Entered and various scans.
- AI can be fed massive amount of past patients data and could help us arrive at a conclusion.
- Can help us minimize the minor errors in learning diagnostics.
- Develop to-the-point drugs, maintain medical records, etc.
- Digital Nurses and Health Bots are future. They’ll provide required consultation.
AI is a vast field and can be confusing on what specific model to use. There have been continuous discussions and multiple methods approached and improvised.
Support Vector Machines
For the purpose of classification and regression, Support Vector Machines can be implemented. Here support vectors are data points, which are closest to the hyperplane. To diagnose the Cancer and other neurological diseases, SVMs are widely used.
Natural Language Processing
We now have large amount of data which is composed of examination results, texts, reports, notes and importantly, discharge information. Now, this data could mean nothing for a machine which has no particular training for reading and learning from such data. This is where NLP could be of use, by learning about keywords related disease and establish a connection with historical data. NLP might have many more applications based on needs.
Implementing hidden layers to identify and establish a connection between input variables and the outcome. The aim is to decrease the average error by estimating the weight between input and output. Image Analysis, Drug Developments and few, are the fields where Neural Networks are harnessed.
As Always, CNNs are the Best:
Convolution Neural Networks, over the time has rapidly being developed and currently is one of the most successful computer vision method. “CNNs simply learns the patterns from the training data set and tries to see such patterns from new images.”. This is same as humans learning something new and implying the knowledge but what all these models know is series of ones and zeros.
With accuracy of 95%, a CNN trained at University of South Florida, can quietly easily detect small lung tumors, often missed by human eye. Another research paper suggests that cerebral aneurysms ca be detected using deep learning algorithms. At Osaka City University Hospital, they detected cerebral aneurysms with 91-93% of sensitivity.
RNNs, which are Recurrent Neural Networks are also popular and could be of great used as they are the neural networks but with information in sequence. Performing a same task for multiple elements and composing output based on last computation.
How Google’s DeepMind sets new milestones:
Acquired by Google in 2014, DeepMind has outplayed many players and has set new record in AI for Health care Industry. Protein Folding is something they have been working on and reached a point where predicting the structure of protein, wholesomely based on its genetic makeup, is possible. What they did was relied on Deep Neural Networks, which are specifically trained to predict protein properties based on genetic sequence. Finally they reached a point where they had the model predict the gap between amino acids and the angles connecting the chemical bonds which connects earlier mentioned amino acids.
This could also help in understanding the underlying reasons of how genetic mutation results in disease. For whenever, the problem with Protein Folding will be solved, it will allows us fasten our process like drug discovery, research, and production of such proteins.
How could this help in aiding Corona (COVID-19)
It is not a new discovery that machine learning can fasten the Drug Development Process for any disease or virus. There are very few datasets available related to Corona and has a lot to tackle in order to establish a conclusion.
Recently, there have been developments involving AlphaFold, which is a computational chemistry related deep learning library.
FluSense using Raspberry Pi and Neural Computing Engine: Started with Lab tests, FluSense is now growing to identify and distinguish human coughing from any other sound, in public places. Idea is to combine the coughing data with people present in the area, which might lead to predict an index of people affected by the flu. This is perfect use case of computer vision in healthcare considering the recent pandemic of covid-19.
Though there have been tremendous developments and many new algorithms are been developed, it would be too early to completely rely on a machine’s output. Efficiently detecting minor diseases around lungs is a great step, but still a small error could lead to catastrophic events. Few more steps towards better models and we can improve the health care, until then we can rely on image analysis systems as an assistant. DataToBiz have been working with few healthcare startups in shaping up their computer vision products/services. Contact our expert to discuss how AI can add value to transform healthcare products/services.