For a machine learning model to help in medical diagnostic tasks, the following features are desired.
1.Good Performance
The algorithm should predict at least as good as a physician . Multinomial Naive Bayes Classifiers are considered as a good classifying algorithm for the purpose. The performance should be so good that the estimations generated should help for further researches.
2.Dealing with missing and noisy data
In medical datasets , very often the description of patients in the records lacks certain data. So the algorithm should be able to handle these cases as well. The data also has many uncertainty and errors. The algorithm should be able to handle noises.
3.Explanation Ability
The system must able to explain decisions when diagnosing new patients so that the physician could understand the results given by the system to confidently test them on the patient.
So, for development of information framework the machine should be fed with the datasets provided. The first task deals with the extraction of all information regarding diseases and treatments. This refined information can be directly provided to the health care providers or the companies that build systematic views. The product can also be developed and sold by companies that do research in medical domain( Machine learning and Natural Language Processing (NLP) ) and companies that develop tools like Microsoft Health Vault and Google Health. The product developed should be trustworthy as it is going to deal with health related problems. For this, the information provided for diseases and treatments needs to based on recent discoveries on health care field.
Firstly, the task is the identification of appropriate sentences from the Medline abstracts which can be done through ML and NLP so that accurate amount of information is extracted. For this extraction there is a major role of the Multinomial Naive Bayes classification algorithm which is used in association with Aprior association rule mining . Further the Weighted Bag-of-word technique helps to consider the frequency and weightage of each feature in the dataset. At the end quality of result is measured with the precision, recall and F-measure.
Conclusion
Thus, machine learning can help in medical field from predicting a diagnosis to a disease to severity of tumors. Machine learning has lots of advantages and can help in Disease identification/diagnosis , personalized treatment, drug discovery/manufacturing, clinical trial research, radiology and radiotherapy, smart electronic health record management, epidemic outbreak prediction for the new diseases.
IBM's Watson AI's success is a live example of what AI can do for cancer treatment. Hence, machine learning and NLP are helping a lot and have a huge scope in further medical discoveries.