Computational modeling and artificial intelligence in disease diagnosis and image analysis in the healthcare system: A review

Abstract

The healthcare industry is going through a significant modernization of its practices, led mostly by the invention and integration of computational modeling and artificial intelligence. Current practices, while effective, are not quick and are often prone to human error. AI can solve the problem of analyzing large datasets and finding patterns by hand due to its immense data processing capabilities. This review stresses the potential value of such AI-powered systems for improving diagnostic accuracy and creating practices with less reliance on arguably cruel approaches to research such as animal testing. In addition, in fields such as medical imaging and cardiovascular care, deep learning models have demonstrated fantastic success, surpassing human-level performance in many cases. This review also highlights the ethical considerations that surround AI itself, such as bias in data and accountability in making critical decisions. The bottom line is that AI and models of computation will change the current landscape of healthcare as emerging AIs and computational models offer faster, more humanized care, turning on a new form of medical options. Simply put, AI and computational modeling will change the scene in healthcare as emerging AIs and models pave a new path towards faster care and a more human-tailored experience. While these innovations are promising if controlled correctly, they also significantly raise ethical concerns, such as the limitations AI will need to have in order to make significant decisions, which is crucial to keeping our health systems humanistic and equitable.

Computational modeling and artificial intelligence in disease diagnosis and image analysis in the healthcare system: A review

Healthcare systems are facing extreme challenges across the globe. One major issue is the increase of cost for many goods used to create products. Another problem is aging global populations leading to a lack of an experienced workforce as well as a higher demand for more products. In addition to these two issues is the overall global increase in diseases and viruses (Bajwa et al., 2021). The current practices in the healthcare industry, while somewhat effective, have been extremely time intensive and the subject of inaccuracies while also lacking in a sense of personalization (Desai, 2020). Additionally, these practices can sometimes be inhumane, as many drugs or products of healthcare have been tested on small animals such as rodents and farm animals (Teerawattananon et al., 2022). In all, the modern practices of the healthcare system have been limited due to the lack of progressive changes into modernizing and virtualizing the industry.

Modern healthcare systems are reliant on the individual prowess of healthcare professionals and the strict compliance to traditional medical protocols (Makimoto & Kohro, 2024). Many things in the healthcare industry are reliant on the judgment or skills of singular healthcare professionals (Ioannidis, 2022). These systems have been effective in recent times, but they are still held back due to the everlasting presence or potential for human error and the cognitive capacities of the human brain. Medical science is a field that is inherently extremely complicated, so if one decision is hastily made or overlooks a minute detail, it could lead to serious consequences. As a result of this, many healthcare systems take time, precision and much collaboration before a decision or product is made, leading to a delay in aiding others, which could be detrimental (Collin et al., 2022).

Recently, the rise of AI (artificial intelligence) has created transformative opportunities to change the healthcare industry for the better (News, 2024). AI has immense potential as it is able to undergo processes and perform tasks effectively and efficiently in many aspects of life (Bajwa et al., 2021). In fact, most of these tasks are not monotonous and would actually require a human level of intelligence, if a human were to do them. AI has the capability of analyzing large datasets and recognizing patterns within the dataset, and then having the ability to make valid predictions. These abilities and the vast unknown of AI have led us to believe that the healthcare industry can be completely revolutionized if it were properly and adequately implemented (Reddy, 2024). Some of these applications include disease diagnosis and predictions where AI can be used to store data to help in predicting certain illnesses or diseases, the development of drugs, as well as the origin of a disease and how to create the most optimal drug to intervene. AI can also be used in medical imaging to analyze and identify certain organs or processes. Figure 1 illustrates some of these applications in healthcare.

Basics of Computational Modeling

Utilizing mainly physics and mathematics, computational modeling is the process of simulating and researching complicated systems. A computational model includes a large number of variables that describe the system under study. The process of simulation involves changing the variables alone or in combination, then observing the results (Collin et al., 2022).

Statistical models use statistics to analyze and make predictions through the patterns of the data (Computational Modeling, NIBIB, 2020). Statistical models are useful in breaking down the relationships present in a larger dataset. However, statistical models have the key flaw that they rely on the assumption that the processes being analyzed can be accurately depicted using conventional probability distributions (Reddy, 2024).

Mechanical models are solely based on the physical properties of the system that is being modeled. The interactions and mechanisms inside the system are described in great depth in these models and the most pertinent disciplines that require mechanical modeling include engineering, chemistry, and physics (Collin et al., 2022).

Basics of AI

Artificial Intelligence (AI) is as the name implies: the intelligence of a human is simulated through computer systems (AI vs. Machine Learning vs. Deep Learning vs. Neural Networks, 2024). Generative AI is artificial intelligence that can make new content strictly by itself. The process of simulating intelligence involves the ideas of learning, where new information causes new boundaries and associations with old information, reasoning, where these large sets of information are used to make conclusions, and self-improvement, where if the computer system is told it did something wrong, it will improve its capabilities (Desai, 2020).

Machine learning (ML) is a type of development in AI where the computer system uses algorithms to analyze and make conclusions based on data (AI vs. Machine Learning vs. Deep Learning vs. Neural Networks, 2024). The main benefit of machine learning is its ability to improve its performance on a certain task with more experience in solving the problem (Rajpurkar et al., 2017).

Similarly to Machine learning, deep learning (DL) is a type of development in AI which uses neural networks, similar to those within the human brain, to visualize complex patterns within a large set of data. Some of these neural networks include RNNs (Recurrent Neural Networks), which are basically used for recognizing patterns in large groups of data, making it useful for long term analysis of a topic or item, GANs (Generative Adversarial Networks), which are used to generate simulated data or images based off prior data, and most importantly, CNNs (Convolutional Neural Networks), which are used to automatically make patterns based off a dataset of images and then easily find anomalies in certain images (Litjens et al., 2017). Deep learning is especially pertinent to Biomedical Engineering because it excels in processing and analyzing vast amounts of medical data, such as medical imaging, genomics, and patient records, which includes medical history, vaccine reports, and lab results (Ardila et al., 2019).

Applications in Disease Prognosis

A specific approach for assessing vaccine effectiveness through computational modeling are cohort studies. These studies involve following two groups of individuals over time, a control group and an experimental group, and comparing their rates of infection. By utilizing statistical models, researchers can adjust for various confounding factors, such as age, sex and socio-economic status, which might influence the results. These models help in understanding the relationship between vaccination and disease occurrence which gives a vast view of the vaccine’s effectiveness (Teerawattananon et al., 2022).

Regression models are extensively used in evaluating vaccine effectiveness. These models can adjust for multiple factors simultaneously which provides a more precise estimate of the vaccine’s effect. Logistic regression is commonly used to analyze binary outcomes like the occurrence of infection, while linear regression might be used for continuous outcomes, such as the severity of symptoms. These models can be further extended to more complex forms like multilevel or hierarchical models, which can account for data that is structured at multiple levels (Ioannidis, 2022).

How has Deep Learning impacted advancements in Disease Diagnosis?

Accurate and timely disease diagnosis is essential in healthcare because it enables treatment plans that work and overall improves patient outcomes. By identifying diseases early, medical professionals can tailor interventions personalized for each patient and reduce the risk of further complications. For example, pneumonia is an infection that causes portions of the lungs to be inflamed. This is often caused by a spread of bacteria, fungi or even viruses. For the most optimal and effective treatment, pneumonia must be accurately diagnosed from an early point in time. Deep learning models have consistently shown impressive results in diagnosing pneumonia.

Created by researchers at Stanford, CheXnet is an impressive deep learning tool that performs very well in detecting pneumonia from chest x-rays. The model was based on a large array of data from ChestX-ray14, which contains over 100,000 x-ray images labeled with certain diseases, one of these including pneumonia. Additionally, the reason CheXnet is considered deep learning is because of the fact the model contains a heavily layered neural network, called the Dense Convulational network, also known as Densenet, which causes the model to learn complex patterns and to acquire an abnormally high rate of accuracy. CheXnet has been extremely accurate, obtaining a pneumonia detection F1 score of 0.435, which is much higher than previous models and even certain radiologists in specificity. The F1 score demonstrates the predictive power an AI model has based on certain tasks rather than overall capabilities (Rajpurkar et al., 2017).

Applications in Medical Imaging

The field of medical imaging has been significantly impacted due to the advancements with the combined integration of traditional methods and AI. Deep learning models have become extremely accurate at diagnosing problems within a patient just from images. Given the context of biomedical engineering, generative AI can help further advancements within medical imaging by making synthetic datasets for training new models given new information. This will address the current problem of limitations on annotated medical images, which will help make better, more efficient AI models in the future. Generative models simulate realistic medical images, which allow researchers and scientists to explore a vast array of different scenarios and to reduce the need for invasive procedures (Artificial intelligence is helping revolutionize healthcare as we know it, 2023).

Lung cancer is the leading cause of death from cancer in the United States as of 2018. Researchers looked into the specific application of DL within lung cancer screenings by using low-dose chest CT scans. By giving the model a 3D convolutional neural network (CNN), the model can analyze and recognize patterns within a great volume of data faster than traditional 2D models which is essential in detecting cancer from an early stage. The model automatically combines multiple different angles of the same scene, which gives it more power when predicting correctly and reducing false positives. All in all, the integration of AI, more specifically Deep Learning, has proven how AI can revolutionize the crucial field of medical imaging (Ardila et al., 2019).

Cardiovascular Imaging

Generative AI and Computational modeling have transformed cardiovascular care by allowing there to be more accuracy when running diagnostics and personalized plans of medication or treatment options. An aspect of cardiovascular care is cardiovascular imaging, which is essential for detecting anomalies early, diagnosing accurately, and monitoring heart diseases, which enables personalized treatment plans and reduces the need for potentially harmful procedures (Khera et al., 2024).

In the article “Transforming Cardiovascular Care With Artificial Intelligence: From Discovery to Practice” by Khera et al., the researchers showed how CNNs have been demonstrating great results when used for cardiovascular imaging. This is because CNNs have remarkable efficiency when identifying patterns inside of images that could be predecessors of worse conditions or even disease. CNNs have been trained on large datasets of echocardiograms, cardiac MRIs, and CT scans to find anomalies like blocked arteries or weaker muscles, competing with human experts in the field.

GANs have been used to simulate and make realistic data of cardiovascular datasets, which have been used to train other AI models. This is especially useful for rarer cardiovascular conditions–such as Brugada Syndrome, which is when the heart’s electrical system is extremely prone to sudden cardiac arrest, or Pulmonary Arterial Hypertension (PAH), which is when the arteries of the lungs and right side of the heart are affected–where gathering real data can be difficult (Makimoto & Kohro, 2024).

Brain Computer Interfaces

Generative AI has made progress in the expanding technology of brain-computer interfaces (BCIs). The innovative technology has enhanced neural signal processing, adaptive systems and personalized care (Eldawlatly, 2024)..

GANs are helping solve neural signal processing by reducing noise and enhancing the clarity of each signal, which allows the devices to be precisely controlled by the owner. Additionally, generative AI allows BCI systems that adapt to adjust in real time for users’ changing neural patterns. This ability to adapt is crucial for maintaining the effectiveness of the BCIs over a long period of time (Eldawlatly, 2024) (Tang et al., 2022).

Bridging the Gap: AI, Computational Modeling, and Healthcare Innovation

The integration of artificial intelligence and computational modeling into healthcare is a great advancement. It not only has transformative benefits and novel applications but also brings new problems from opportunities that arise from this achievement. This section discusses its strengths and weaknesses, the ethical concerns AI could bring with it and where in the future this area might develop.

AI has already demonstrated an immense amount of capabilities in increasing the efficiency and effectiveness of healthcare, as seen in Figure 1 with its immense amount of applications. It is already quite advanced in comparison to human capability, as it allows analysis of large sets of data and makes conclusions and predictions from them. This capability to predict from data is not only especially useful in fields like disease diagnosis, but also medical imaging and cardiovascular care. Its ability to process millions of images and inform where the potential threats are is almost human: even more accurate than experienced professionals as Chexnet experiments demonstrated (Desai, 2020) (Rajpurkar et al., 2017).

The issue at hand is the point to which AI can be extended in healthcare. As an example, one challenge is the fact that models are trained on credible data. Obtaining this data is particularly difficult for rare diseases. Moreover, models built on AI still have errors to deal with. This is demonstrated in the cases where models are looking at data that they have not seen before, and it demonstrates the need for further development and supervision by humans over AI models (Collin et al., 2022).

These applications of AI in healthcare also raise important ethical issues that must be addressed to ensure their responsible deployment. AI algorithms are subject to bias if the training data do not cover all characteristics of the population. Thus, AI models can unintentionally support disparities and biases in healthcare. Furthermore, replacing medically trained personnel with AI in the decision-making process may dehumanize care. To combat these concerns, there is a need for maximum transparency and supervision when AI is used for all critical decisions (Bajwa et al., 2021)..

On the other hand, the use of animals in drug testing and new healthcare product development is highly criticized as inhumane. AI-driven models and simulations are, thus, a suitable option to remove the need for animal testing, making AI a considerably ethical decision when conducting biomedical research (News, 2024). Additionally, the FDA has recently been promoting the usage of computational tools as they are extremely effective and combat the usage of animal testing. (Commissioner, 2024).

In the future, AI is expected to play an even greater role in the healthcare field. Advancements in generative AI and computational modeling will likely allow for more personalized and precise treatments. For example, AI-mediated BCIs will transform care by creating an adaptive system that changes according to a user’s neural patterns. Similarly, personalization can be used when designing or prescribing certain drugs for different patients. Growing areas of AI development such as AI-assisted medical imaging will also improve diagnosis and in turn reduce the need for invasive diagnostic procedures (Reddy, 2024).

As AI and computational modeling merge in health care, diagnostic accuracy improves and personalization of treatments becomes more powerful. The advancement of AI also allows for less dependence on practices like animal testing. But these advancements also bring their own set of problems, such as the need for data, possible biases in results and how best to handle ethical issues that arise along with them. As AI moves forward, it will have an increasing amount of applications not only in medical settings, but also in robots that can conduct surgeries or devices that monitor patients from home. These devices will communicate directly with trained medical professionals virtually. We can make the most of these innovations to produce more humane, higher-quality individualized patient service. The future of healthcare lies in the merging of AI, computational modeling and human expertise for a more efficient and less need for human expertise, reducing human error in healthcare overall (Artificial intelligence is helping revolutionize healthcare as we know it, 2023).

References

Accuracy of a Generative Artificial Intelligence Model in a Complex Diagnostic Challenge | Artificial Intelligence | JAMA | JAMA Network. (n.d.). Retrieved August 4, 2024, from https://jamanetwork.com/journals/jama/fullarticle/2806457

AI vs. Machine Learning vs. Deep Learning vs. Neural Networks | IBM. (2024, April 15). https://www.ibm.com/think/topics/ai-vs-machine-learning-vs-deep-learning-vs-neural-net works

Angermueller, C., Pärnamaa, T., Parts, L., & Stegle, O. (2016). Deep learning for computational biology. Molecular Systems Biology, 12(7), 878. https://doi.org/10.15252/msb.20156651

Ardila, D., Kiraly, A. P., Bharadwaj, S., Choi, B., Reicher, J. J., Peng, L., Tse, D., Etemadi, M., Ye, W., Corrado, G., Naidich, D. P., & Shetty, S. (2019). End-to-end lung cancer screening with three-dimensional deep learning on low-dose chest computed tomography. Nature Medicine, 25(6), 954–961. https://doi.org/10.1038/s41591-019-0447-x

Artificial intelligence is helping revolutionize healthcare as we know it. (2023, September 14). JNJ.Com. https://www.jnj.com/innovation/artificial-intelligence-in-healthcare

Bajwa, J., Munir, U., Nori, A., & Williams, B. (2021). Artificial intelligence in healthcare: Transforming the practice of medicine. Future Healthcare Journal, 8(2), e188–e194. https://doi.org/10.7861/fhj.2021-0095

Collin, C. B., Gebhardt, T., Golebiewski, M., Karaderi, T., Hillemanns, M., Khan, F. M., Salehzadeh-Yazdi, A., Kirschner, M., Krobitsch, S., & Kuepfer, L. (2022). Computational Models for Clinical Applications in Personalized Medicine—Guidelines and Recommendations for Data Integration and Model Validation. Journal of Personalized Medicine, 12(2), 166. https://doi.org/10.3390/jpm12020166

Commissioner, O. of the. (2024). Modeling & Simulation at FDA. FDA. https://www.fda.gov/science-research/about-science-research-fda/modeling-simulation-fd a

Computational Modeling. (n.d.-a). Glenn Research Center | NASA. Retrieved July 23, 2024, from https://www1.grc.nasa.gov/space/human-research-program/computational-modeling/

Computational Modeling. (2020). National Institute of Biomedical Imaging and Bioengineering Retrieved July 30, 2024, from https://www.nibib.nih.gov/science-education/science-topics/computational-modeling

Desai, A. N. (2020). Artificial Intelligence: Promise, Pitfalls, and Perspective. JAMA, 323(24), 2448–2449. https://doi.org/10.1001/jama.2020.8737

Eldawlatly, S. (2024). On the role of generative artificial intelligence in the development of brain-computer interfaces. BMC Biomedical Engineering, 6(1), 4. https://doi.org/10.1186/s42490-024-00080-2

Ioannidis, J. P. A. (2022). Factors influencing estimated effectiveness of COVID-19 vaccines in non-randomised studies. BMJ Evidence-Based Medicine, 27(6), 324–329. https://doi.org/10.1136/bmjebm-2021-111901

Litjens, G., Kooi, T., Bejnordi, B. E., Setio, A. A. A., Ciompi, F., Ghafoorian, M., van der Laak, J. A. W. M., van Ginneken, B., & Sánchez, C. I. (2017). A survey on deep learning in medical image analysis. Medical Image Analysis, 42, 60–88. https://doi.org/10.1016/j.media.2017.07.005

Makimoto, H., & Kohro, T. (2024). Adopting artificial intelligence in cardiovascular medicine: A scoping review. Hypertension Research, 47(3), 685–699. https://doi.org/10.1038/s41440-023-01469-7

News, N. (2024, May 10). A New Era in Neuroscience with Generative AI. Neuroscience News. https://neurosciencenews.com/neuroscience-generative-ai-26085/

One step closer to a brain-computer interface: Generative AI decodes images and speech from fMRI recordings. (n.d.). Retrieved August 11, 2024, from https://www.from-the-interface.com/generative-ai-fmri/

Rajpurkar, P., Irvin, J., Zhu, K., Yang, B., Mehta, H., Duan, T., Ding, D., Bagul, A., Langlotz, C., Shpanskaya, K., Lungren, M. P., & Ng, A. Y. (2017). CheXNet: Radiologist-Level Pneumonia Detection on Chest X-Rays with Deep Learning (arXiv:1711.05225). arXiv. https://doi.org/10.48550/arXiv.1711.05225

Reddy, S. (2024). Generative AI in healthcare: An implementation science informed translational path on application, integration and governance. Implementation Science : IS, 19, 27. https://doi.org/10.1186/s13012-024-01357-9

Rohan Khera, M. D., Evangelos K. Oikonomou, M. D., Girish N. Nadkarni, M. D., Jessica R. Morley, P., Jenna Wiens, P., Atul J. Butte, M. D., & Eric J. Topol, M. D. (2024). Transforming Cardiovascular Care With Artificial Intelligence: From Discovery to Practice: JACC State-of-the-Art Review. Journal of the American College of Cardiology. https://doi.org/10.1016/j.jacc.2024.05.003

Takagi, Y., & Nishimoto, S. (2023). High-Resolution Image Reconstruction With Latent Diffusion Models From Human Brain Activity. 14453–14463. https://openaccess.thecvf.com/content/CVPR2023/html/Takagi_High-Resolution_Image_ Reconstruction_With_Latent_Diffusion_Models_From_Human_Brain_CVPR_2023_pap er.html

Tang, J., LeBel, A., Jain, S., & Huth, A. G. (n.d.). Semantic reconstruction of continuous language from non-invasive brain recordings.

Teerawattananon, Y., Anothaisintawee, T., Pheerapanyawaranun, C., Botwright, S., Akksilp, K., Sirichumroonwit, N., Budtarad, N., & Isaranuwatchai, W. (2022). A systematic review of methodological approaches for evaluating real-world effectiveness of COVID-19 vaccines: Advising resource-constrained settings. PLOS ONE, 17(1), e0261930. https://doi.org/10.1371/journal.pone.0261930

---