On March 3rd, 2010, H.R.6216 - National Artificial Intelligence Initiative Act of 2020 was introduced to the House as an initiative for the federal government to play a more significant role in investing, researching, and developing artificial intelligence in all sectors of our American society. This initiative serves to:
“To establish the National Artificial Intelligence Initiative, and for other purposes….(that) provide sufficient resources and use its convening power to facilitate the growth of artificial intelligence human capital, research, and innovation capacity in academia and other nonprofit research organizations, companies of all sizes and across all sectors, and within the Federal Government.” (116th Congress 2D Session)
Recently artificial intelligence has shown significant promise in advancing research and applications in the medical field which we will outline from recent studies.
AI and Drug Discovery
Computational models using AI deep learning has shown great potential for analyzing properties of chemicals from a big database for prospective processes and activities in drug discovery research. The more complex the property being analyzed in the model, the less reliable and predictive the model performs modeled by figure 1 .
Several obstacles must further be developed to more reliably apply deep learning models to analyze big data for drug discovery with complex mechanisms such as predicting animal responses and human efficacy and side effects. Some of the obstacles to working with big data in this field is the scale of the data, the continued growth or input of the data, the variety in sources contributing to the data, and the uncertainty of the data . Missing data is also a regular issue that arises . When the property or mechanism is simple, a model can be developed for the individual analysis of potential compounds and use the resulting models to predict compounds not tested . However, deep learning and data sets need to be improved to advance the use of AI for complex mechanisms.
AI has come a long way from the its birth in the 1950s and the first attempts at computer aided drug discovery in the 1970s. As hardware improved in storage capacity and processing power, construction of artificial neural networks, computing and analysis of big data became a realization, giving way to deep learning requiring for computing the more complex mechanisms and properties in the data we have available today shown in figure 3 .
Other AI Applications in Medicine
AI has also been applied to nanotechnology for modeling prospective designs for new nanomaterials (Zhu, 2020). The field has great potential to expand however is limited by the chemical descriptor database (Zhu, 2020). Nanomaterials have a variety of properties and complex structural make up which causes molecular dynamic simulations to be expensive to compute (Zhu, 2020). Another AI application is the use of CNN for image modeling. Inspired by neuroscience, CNN has been used to detect cancer, Alzheimer’s disease, and heart disease (Zhu, 2020). CNN has also been used to predict cancer outcomes from tissue images and genetic markers and recognize images to predict molecular interactions .
COVID-19 Detection with AI
One research study constructed a model for AI detection of COVID-19 on X-rays, work radiologists perform currently under intense stress. In this study 26o images, 130 of COVID-19 and 130 normal, were selected from a public database and 30 were selected at random to test against a deep learning convolutional neural network (CNN) model . This model achieved 100% sensitivity, specificity, accuracy, PPV, and NPV . The results determine that this model could produce results that will decrease reading time for radiologists and perform at the expert level of a radiologist which would have the potential to improve early diagnosis and help to control the epidemic .
These data points from this study seems to allude to a possible replacement to the risk of human error. With 100% efficiency in the measured variables, why continue to use human radiologist? Another study incorporated other variables that human radiologists observe when performing this work in the larger scope of cases they analyze.
The dataset analyzed by this model consisted of 4356 chest CT exams from 3,322 patients . Community acquired pneumonia (CAP) and other non-pneumonia CT exams were included to test the efficiency of the model under a multitude of variables beyond the normal . The model achieved 90% sensitivity and 96% specificity in detecting COVID-19, 87% sensitivity and 92% specificity in detecting CAP, and 94% sensitivity and 96% specificity in detecting non-pneumonia in the CT exams . Due to the similarities in lung response with a variety of illnesses, images will produce similar results on a chest CT. common misclassification.
This data shows promise for the potential of using this AI model to differentiate COVID-19 from other respiratory illnesses. However, due to the margin of error, the current model results could potentially serve as a prescreening in high pressure environments but would still require human intelligence to verify the diagnosis. More work would have to be done to improve the sensitivity and specificity for this technology be work as a stand-alone solution for COVID-19 detection.
Drug Treatment and Personalized Medicine
Treatment regimens and drug administration strategies have become more complicated as more knowledge is acquired about how chemical properties, patient medical information, genetic information, and chemical interactions determine outcomes. Predicting combination therapies for cancer treatment and infectious disease by using AI to analyze data to determine drug interactions with chemical properties and biological activities has been a developing field to personalizing medicine for greater efficacy . As administration of drugs continue to advance such as slow release mechanisms for internally embedded drugs for birth control, diabetes treatment, or cancer treatment, regulating the concentration of the drug in the body to maintain an effective dosage below the toxicity level for the body will need to be effectively computed. AI serves as a promising model for predicting interactions to better compute the best concentration to minimize side effects .
With the current capacity of the human brain, we will never be able to access and utilize every bit of information regarding chemicals, biological processes, and patient information for every drug treatment administration on our own. The need for AI is based on the ability to simplify the data to a package that can be processed and utilized by a human professional with better efficiency than our human brain can perform in the given time.
“There is no AI that will function as a skeleton key, recommending treatments for all diseases. Instead, tools are being developed to handle specific tasks in the drug selection and administration process of each disease and each patient” (Room & Tsigelny, 2020)
1. Salman, F.M., Abu-Naser, S. S., Alajrami, E., Abu-Nasser, B. S., & Alashqar, B. A. M. (2020). COVID-19 Detection using Artificial Intelligence. Retrieved From: http://dstore.alazhar.edu.ps/xmlui/handle/123456789/587
2. Li, L., Qin,L., Xu, Z., Yin, Y., Wang, X., Kong, B., Bai, J., Lu, Y., Fang, Z., Song, Q., Cao, K., Liu, D., Wang, D., Xu, Q., Fang, X., Zhang, S., Xia, J., Xia, J. (2020). Artificial Intelligence Distinguishes COVID-19 from Community Acquired Pneumonia on Chest CT. Retrieved from: https://pubs.rsna.org/doi/full/10.1148/radiol.2020200905
3. Zhu, H. (2020). Big Data and Artificial Intelligence Modeling for Drug Discovery. Annual Review of Pharmacology and Toxicology. 60:573–89. Retrieved from: https://www.annualreviews.org/doi/10.1146/annurev-pharmtox-010919-023324
4. Romm, E. L., and Tsigelny, I.F. (2020). Artificial Intelligence in Drug Treatment. Annual Review of Pharmacology and Toxicology. 60:353–69. Retrieved from: https://www.annualreviews.org/doi/10.1146/annurev-pharmtox-010919-023746
Geralt. Artificial intelligence brain think control. Retrieved from: https://pixabay.com/illustrations/artificial-intelligence-brain-think-3382507/