Application of Deep learning in healthcare

With advanced computing technology along with growth in big data across the globe, there is a growing demand in the use of Deep Learning. Deep learning can be defined as a sub-set of Artificial Intelligence (AI). Deep Learning has been extensively been used in medical applications like anatomic modelling (segmenting of anatomic structures), tumor detection, disease classification, computer-aided diagnosis and surgical planning. (Esteva, 2019).  Data is continuously been generated in biomedical research that ranges from electronic health records, imaging, -omics, sensor data and text. And these data are mainly unstructured, poorly annotated, complex and heterogenous. Hence, traditional data mining techniques are inefficient in processing these data. (Miotto, 2018)

Machine Learning technique is a form of computer programming that that transforms the inputs of an algorithm into useful outputs based on statistical, data-driven rules that are automatically extracted from large datasets rather than supervised by humans. And hence, it requires expert supervision. On the other hand, Deep learning extracts its patterns and features from a set of raw data, without any supervision being required. Hence, it requires multiple layers of representation.  (Esteva, 2019) This is the fact that makes Deep Learning different from Machine Learning.

Deep learning systems are capable of assisting physicians by offering second opinion and indicating concerning areas in the image. Convolution Neural Networks (CNN) is a deep learning algorithm that can process data that exhibits natural spatial invariance, such as Images (that do not change its meaning under translation). This algorithm is widely used in medical imaging in complex diagnostics spanning pathology, dermatology, ophthalmology and radiology. Recurrent Neural Network (RNN) is another deep is another deep Learning technique that is widely used in healthcare. This deep learning technique is effective in processing sequential data inputs specially speech, language and time-series data. This technique can be exclusively used in processing EMRs, which is a rich source of unexplored data in medical field. (Esteva, 2019)

Deep learning models have achieved physician -level accuracy in a wide range of diagnostic tasks such as diabetic retinopathy, mole identification from melanomas, referrals from fundus, cardiovascular risks and so on. (Esteva, 2019)

Researchers mainly depends on unsupervised learning such as auto-encoders, when dealing with EMRs. In this approach networks are first trained to learn from an unlabeled set of data by compressing and reconstructing the data. Most of these works focus on MIMIC (Medical Information Mart for Intensive care (MIMIC) dataset that consist of medical records on ICU (Intensive Care Unit) patients from a single medical canter. As ICU patients generate high volume data compared to non-ICU patients. (Esteva, 2019)

Deep Learning has scope of creating next generation predictive health care system, that is capable of assisting physicians in clinical decisions by referring medical records of billions of patients. This technique also has the potential to support hypothesis-driven research and exploratory investigation in medical field. (Miotto, 2018)

Reference:

Esteva, A., Robicquet, A., Ramsundar, B., Kuleshov, V., DePristo, M., Chou, K., Cui, C., Corrado, G., Thrun, S. and Dean, J., 2019. A guide to deep learning in healthcare. Nature medicine, 25(1), pp.24-29.

Miotto, R., Wang, F., Wang, S., Jiang, X. and Dudley, J.T., 2018. Deep learning for healthcare: review, opportunities and challenges. Briefings in bioinformatics, 19(6), pp.1236-