Applications of Deep Learning in Health Informatics

Author(s): Ankit Sirmorya

Originally published on Towards AI the World’s Leading AI and Technology News and Media Company. If you are building an AI-related product or service, we invite you to consider becoming an AI sponsor. At Towards AI, we help scale AI and technology startups. Let us help you unleash your technology to the masses.

Deep Learning

Introduction

The deep learning (DL) computing paradigm has gained popularity within the machine learning (ML) community over the last few years. The computational model has also gradually become the most common computational approach in the field of ML, so as to achieve exceptional results on several complex cognitive tasks, matching or even beating those provided by humans[9]. The power of DL is its ability to learn massive amounts of data. The role of data analytics in health informatics has grown rapidly over the last decade as a result of a massive influx of multimodality data. Health informatics has also seen an increase in interest in using machine learning to generate analytical, data-driven models [4]. In the past few years, it has become apparent that Medical Imaging and Public Health are increasingly using Deep Learning. Fig 1 illustrates an explosion of interest in deep learning over the past few years as indicated by the number of publications in the subfields of medical imaging, medical informatics, and pervasive sensing [4].

In the healthcare sector, there is quite a lack of exploration, as healthcare professionals work round the clock to keep their patients alive and provide them with the best possible treatment. Taking Covid-19 as an example, health care workers have worked tirelessly to identify the infection through Chest X-rays, PCR tests on blood or sputum samples, etc. As the number of cases grew, it became very important to develop solutions that could predict possible outcomes based on the data at hand. In an effort to control the growing numbers of infections, Deep Learning models were used to develop such solutions [1]. With the 2020 outbreak of the novel coronavirus (COVID-19) [2], DLs will have an increasingly critical role to play in early diagnosis. In recent years, CNN has succeeded in taking some medical imaging classification and localization tasks to new levels of accuracy and precision, going beyond traditional diagnosis to automated diagnosis. In contrast to conventional machine-learning and data mining techniques, deep learning can generate very high-level data representations from massive amounts of data. A number of reasons have led to this widespread adoption of DL, among them:

(i) DL is robust to changes in input data because we do not need precisely designed features and we learn optimized features automatically

(ii) It is possible to generalize DL techniques in different applications by using a method known as transfer learning, which utilizes knowledge gained from solving one problem to solve another related problem

(iii) The DL techniques are highly scalable.

Benefits of Deep Learning in Health Sciences

While Deep Learning has a lot of potential in Health Sciences, the power of Deep Learning lies in the data that is going to be used to train the models, and to do so, we need to have enough data to train the models and get better prediction. As Deep Learning applications go beyond the traditional modeling and prediction of something, it also has more potential to be able to identify and understand certain patterns.

According to a study, Convolutional Neural Networks detected Melanoma with up to 10 percent more accuracy and specificity than human physicians [7]. A number of benefits are found to be associated with Deep Learning that could be beneficial to the healthcare system as a whole.

Making accurate diagnostic suggestions with deep learning algorithms can:

1. Reduce healthcare costs.
2. Prevent critical and urgent cases from being delayed.

3. Healthcare professionals can focus on more complex diagnostics or have more time to care for patients if administrative burdens are reduced

4· By auditing diagnostic procedures and prescriptions, errors in diagnosis can be reduced

5· Leading to faster diagnosis

Use Cases of Deep Learning on Health Sciences

This section describes how Deep Learning has proven to be very useful in the domain of healthcare. We will discuss different use cases of Deep Learning in this section.

Medical Imaging

When it comes to finding out and diagnosing a certain issue, medical imaging is key. Image recognition is being used to identify or detect diseases and highlights problems in magnetic resonance images, X-rays, and other images [8]. Deep Learning models are intelligent enough to detect patterns within an image. With Convolutional Neural Networks and Deep Neural Networks, which require a lot of data to train well enough to provide accurate predictions, these models identify and learn certain patterns that allow them to identify such patterns in real-world situations with sufficient intelligence.

Patient Care Data Analytics

As a result of current tools and technologies, it is much easier to perform in-depth analyses of the data available in the form of Electronic Health Records. Keeping track of these records is helpful in identifying possible problems that may arise. Using Electronic Health Records, for instance, Adverse Drug Events may be identified to determine previous medications that could cause such events in the future. Deep Learning models are also used in the field of patient care data analytics to identify potential patterns and present them as meaningful insights [6].

Health Assistants

There are health assistant chatbots that aim to provide information that is true and useful to the patient, such as symptoms-based diagnosis or chatbots that help people overcome mental disorders. Our Chatbots can be trained to utilize Deep Learning to understand the context of the conversation as well as provide useful feedback to the user. LSTM (Long Short-Term Memory) models are neural networks that are used to develop chatbots.

Deep Learning has many potential uses in health sciences such as personalized treatments, genome analysis, and prescription audits.

Popular Deep Learning Architectures

There are several different Deep Learning models which could be used to solve different problems, so here is a look at some of the most common Deep Learning architectures which could be used in the health science field. Because neural networks are inspired by the brain, the idea of Perceptron hovers across the different architectures [Table 1] of Deep Learning models. As a method of improving the Network’s performance, hidden layers consisting of Perceptrons are added, which act as an intermediary between the input neurons and the output neurons that provide the outputs. In order to solve more complex problems, the model is trained over a number of epochs during which it learns the patterns of identification based on input data [4].

Deep Learning Methods by areas and applications in health informatics

Deep Learning consists of a wide variety of algorithms that are a combination of several different algorithms and neural networks. With the architectures that we have discussed above in mind, let us explore some of the potential use cases that could be implemented with these Deep Learning Architectures. Figure 2 illustrates a pattern of publications for deep learning methods since 2010[4]. In recent years, CNNs have had a growing impact on the field of health informatics.

Medical Imaging

Medical imaging data can be used to solve problems such as Neural Cells Classification, Tissue Classification, and Organ Segmentation using Convolutional Neural Networks. An MRI scan, CT scan, or endoscopy result can be used as input data. By having such data at hand, we could build more intelligent and precise models.

Bioinformatics

Bioinformatics has many potential applications including cancer diagnosis, gene selection, drug design, etc. Using Deep Auto-encoder and Deep Neural Networks, it may be much easier to diagnose or detect cancer-based on gene expression data. In this domain, there is a larger space that needs to be captured, and this can be done with the help of Deep Belief Networks, Auto-encoders, and Neural Networks.

Medical Informatics

Medicinal Informatics has huge potential by addressing issues like disease prediction, monitoring human behavior, and data mining using data such as Electronic Medical Records, Blood Tests, etc. With Deep Belief Networks, Convolutional Neural Networks, etc., these problems could be addressed more accurately.

Among the other areas in Health Informatics where deep learning can be applied are Public Health (Predicting demographic information, Lifestyle diseases, Infectious disease epidemics) and Pervasive Sensing (Anamoly detection, Human activity recognition, Obstacle detection, etc.). The table below summarizes the different applications in each of the five areas of health informatics.

Deep Learning in healthcare: Limitations and challenges

Deep Learning models have evolved and progressed significantly since they were first developed. As new technology and models emerge every year, there has been a lot of concern about how accurate they are or if there are any limitations where they won’t work. There are also problems with interpreting Deep Learning Models, and people sometimes use them as black boxes. Another common limitation is not having enough data to train a good network and what is more important than having enough data is having the right data, as sometimes the slightest change in data can cause the network to give false results. Over time, more and more data will be generated, which will allow us to train more intelligent and efficient networks in the future to become totally dependent on them [4].

Conclusion

Throughout history, health sciences have achieved major breakthroughs and have progressed towards a technological age. Several technologies are available, including Data Science, Artificial Intelligence, Machine Learning, Deep Learning, and Natural Language Processing. Through the provision of accurate predictions, more in-depth analysis of the problem, and accurate doses of medicine, these technologies have the potential to revolutionize the healthcare sector. As these features become more and more real, doctors are now beginning to trust predictions and results provided by a machine learning model, or we can say that it is assisting them in taking better decisions.

In healthcare, deep learning is still at its very early stages and needs more time to develop and improve efficiency in order to provide better and more precise results. The use of Deep Learning and Data Analytics in healthcare is more prevalent today than ever before. Deep Neural Networks are still dependent on how they are trained and on their ability to predict any outcome. The use cases we reviewed are all human-dependent and more than that, they are data-dependent, which we do not have much of right now. When these technologies are fully integrated into the healthcare system, they will be able to predict much more accurately, and at that point, people can completely rely on them.

As it is more precise, accurate, deep learning has the potential to improve the healthcare industry. Due to the increasing amount of work healthcare professionals have to handle every day, it has become more and more necessary to understand the importance of such technologies. AIDoc [5] is a platform that provides tools for healthcare professionals in order to make their jobs easier. Keeping the extra burden off of them will allow them to focus on what really matters.

References:

1. Brunese L, Mercaldo F, Reginelli A, Santone A. Explainable Deep Learning for Pulmonary Disease and Coronavirus COVID-19 Detection from X-rays. Computer Methods and Programs in Biomedicine. 2020 Nov 1;196:105608.

2. Jamshidi M, Lalbakhsh A, Talla J, Peroutka Z, Hadjilooei F, Lalbakhsh P, et al. Artificial Intelligence and COVID-19: Deep Learning Approaches for Diagnosis and Treatment. IEEE Access. 2020;8:109581–95.

3. Hoo-Chang Shin, Orton MR, Collins DJ, Doran SJ, Leach MO. Stacked Autoencoders for Unsupervised Feature Learning and Multiple Organ Detection in a Pilot Study Using 4D Patient Data. IEEE Trans Pattern Anal Mach Intell. 2013 Aug;35(8):1930–43.

4. Ravì D, Wong C, Deligianni F, Berthelot M, Andreu-Perez J, Lo B, et al. Deep Learning for Health Informatics. IEEE Journal of Biomedical and Health Informatics. 2017 Jan;21(1):4–21.

5. Deep Learning in Healthcare and Radiology | Aidoc Blog [Internet]. Aidoc. 2020 [cited 2021 Dec 21]. Available from: https://www.aidoc.com/blog/deep-learning-in-healthcare/

6. Deep Unsupervised Learning for Healthcare Data Analytics | Hindawi [Internet]. [cited 2021 Dec 21]. Available from: https://www.hindawi.com/journals/jhe/si/858632/

7. HealthITAnalytics. Deep Learning Tool Tops Dermatologists in Melanoma Detection [Internet]. HealthITAnalytics. 2018 [cited 2021 Dec 21]. Available from: https://healthitanalytics.com/news/deep-learning-tool-tops-dermatologists-in-melanoma-detection

8. 11 Deep Learning Use Cases / Applications in Healthcare in 2021 [Internet]. 2021 [cited 2021 Dec 21]. Available from: https://research.aimultiple.com/deep-learning-in-healthcare/

9. Alzubaidi, L., Zhang, J., Humaidi, A.J. et al. Review of deep learning: concepts, CNN architectures, challenges, applications, future directions. J Big Data 8, 53 (2021). https://doi.org/10.1186/s40537-021-00444-8

---

Applications of Deep Learning in Health Informatics was originally published in Towards AI on Medium, where people are continuing the conversation by highlighting and responding to this story.

Join thousands of data leaders on the AI newsletter. It’s free, we don’t spam, and we never share your email address. Keep up to date with the latest work in AI. From research to projects and ideas. If you are building an AI startup, an AI-related product, or a service, we invite you to consider becoming a sponsor.

Published via Towards AI