The convergence of precision medicine (PM) and artificial intelligence (AI) can significantly transform the healthcare sector. PM methods can identify phenotypes of patients with unique healthcare needs or less common responses to treatment, while AI can leverage sophisticated computation and inference for generating insights, enable systems to learn and reason and empower clinical decision-making through augmented intelligence. This article discusses the role and application of AI methods in PM.
Image credit: PopTika/Shutterstock
Importance of AI in PM
In clinical research and patient care, PM is an emerging approach that focuses on treating and understanding diseases by integrating multi-omics/multi-modal data from an individual to make patient-tailored decisions. The application of PM diagnostic approaches led to the generation of complex and large datasets, which can be processed and understood effectively using novel techniques.
AI techniques, such as machine learning (ML), can identify complex patterns within data that can be utilized to make classifications/predictions on new data or for advanced exploratory data analysis. ML analysis of PM’s multi-modal data enables extensive investigation of large datasets, eventually leading to a better understanding of human disease and health.
Diagnosis and Disease Prediction: Integrating AI with PM can significantly improve the effectiveness, speed, and accuracy of diagnosing diseases, leading to better patient outcomes. AI algorithms can efficiently process and analyze substantial amounts of medical data, including diagnostic images and patient data, to identify subtle correlations and patterns that are not easily detectable through human observation.
For instance, recent studies have demonstrated that an AI model can outperform human dermatologists in diagnosing skin cancer accurately. The AI model trained using a dataset containing more than 130,000 images attained 95% accuracy compared to 86.6% achieved by human dermatologists. This indicates the feasibility of using AI to enhance diagnostic capabilities and provide crucial insights to healthcare professionals for timely and accurate diagnoses.
AI can also be used in PM to assess risk factors and predict diseases. AI algorithms can analyze large datasets consisting of patient data, medical records, and genetic profiles to detect patterns that indicate the possibility of developing specific diseases.
For instance, an AI algorithm has been developed that can predict the onset of Alzheimer’s disease several years before the manifestation of symptoms. The algorithm realized a high accuracy of 82% in identifying individuals who would develop Alzheimer’s at a later stage by analyzing the patient data and brain scans, indicating AI's potential in early disease prediction that enables personalized preventive strategies and proactive interventions.
AI can play a crucial role in real-time patient health monitoring, which is a key aspect of PM. AI algorithms can constantly analyze the incoming data streams generated by health monitoring platforms, wearable devices, and Internet of Things (IoT) sensors to detect anomalies, monitor vital signs, and provide meaningful insights to healthcare professionals.
For instance, AI-driven monitoring systems can analyze physiological parameters such as blood pressure and heart rate variability to identify early signs of cardiac arrhythmias or abnormalities and alert healthcare providers in real time, which can ensure the initiation of immediate interventions, leading to improved patient outcomes and saving the lives of at-risk individuals.
Moreover, AI can analyze patient data from several sources, such as patient-reported outcomes, genomic data, and electronic health records (EHRs), to generate comprehensive insights into the health status of a patient, which allows healthcare professionals to make more informed decisions on personalized interventions and treatments.
Drug Discovery and Development: AI-powered drug discovery utilizes computational methods and algorithms to design, develop, and identify new drugs. AI algorithms can analyze large datasets containing chemical and biological data to enable researchers to identify potential drug candidates more effectively and efficiently.
For instance, biotechnology company Insilico Medicine has employed AI to identify a novel drug candidate for idiopathic pulmonary fibrosis. The company leveraged deep learning (DL) algorithms to screen millions of compounds to identify a molecule with robust anti-fibrotic properties, significantly reducing the resources and time required for drug discovery compared to traditional methods.
AI can also identify the potential drug targets, which are primarily specific cellular processes or molecules involved in disease pathways. AI algorithms can reveal unknown targets for therapeutic intervention by analyzing large-scale genomic and biological data.
For instance, The Cancer Genome Atlas (TCGA) project analyzed molecular and genomic data from several cancer patients using AI to identify specific genetic alterations related to different cancer types, leading to the discovery of new drug targets, such as over-expressed proteins and specific mutations, which can be targeted by PM.
Moreover, AI-powered technologies can optimize drug safety and efficacy by analyzing large datasets and predicting drug treatment outcomes. AI algorithms can personalize treatment regimens to minimize side effects and maximize efficacy by considering multiple factors, such as treatment response data, disease characteristics, and patient genetics.
For instance, a recent study demonstrated that AI algorithms can be used effectively for drug-induced cardiac toxicity prediction, a major concern during drug development. The AI system predicted cardiac toxicity with a high accuracy of over 90% by analyzing clinical data and molecular structures.
AI Techniques Used in PM
Support Vector Machine (SVM): SVM can analyze and classify symptoms for better diagnostic accuracy. Additionally, SVM can be utilized to identify biomarkers of psychological and neurological diseases and analyze single nucleotide polymorphisms (SNPs) to validate breast cancer and multiple myeloma.
SVM can also analyze pathological, epidemiological, and clinical data to prevent cervical and breast cancer and molecular, genomic, and clinical data to diagnose mental disease and validate oral cancer.
DL: DL is used extensively to analyze images from several healthcare sectors, specifically oncology. DL can be employed to analyze lung cancer, computed tomography (CT) scans, magnetic resonance imaging (MRI) of the pelvic and abdominal area, mammography, brain scan, biomarker data and sequencing, glioma through histopathological scanning, and radiographs of malignant lung nodules.
Moreover, DL can also be applied in the diagnostic process of diabetic retinopathy, histopathological anticipation in women with cytological deformations, nodular basal cell carcinoma, dermal nevus and seborrheic keratosis, cardiac muscle failure, and cardiac abnormalities.
Decision Tree (DT): DT is primarily used for real-time healthcare monitoring, therapeutic decision support systems, and detecting sensor aberrant data. Real-time application of the DT algorithm includes therapeutic decision-making in psychological patients, identification of health outcome predictors, diagnosing hypertension through finding factors, supporting clinical decisions, locating genes associated with pressure ulcers (PUs) among elderly patients, and finding the potential telehealth services for patients.
Random Forest (RF): RF is utilized to identify nonmedical factors related to health, classify and diagnose Alzheimer’s disease, diagnose mental illness, identify factors associated with diabetic peripheral neuropathy diagnosis, and predict metabolic pathways of individuals, mortality of intensive care unit (ICU) patients, disease risks from clinical error data, healthcare cost, and results of a patient’s encounter with the psychiatrist.
Naïve Bayes (NB): NB is employed to classify EHR, shape the clinical diagnosis for decision support, extract genome-wide data for Alzheimer's disease identification, model a cardiovascular disease-related decision, measure the quality of healthcare services, predict risks by identifying Mucopolysaccharidosis type II, and construct a predictive model for cancer in prostate, breast, and brain.
Genetic Algorithm (GA): GA can be used in infectious disease, neurology, orthopedics, pediatrics, cardiology, pulmonology, surgery, gynecology, oncology, radiology, and endocrinology. Other AI techniques used in PM include the hidden Markov model (HMM), linear regression, discriminant analysis, logistic regression, and k-nearest neighbor (KNN).
Recent Studies
In a recent study in Frontiers in Artificial Intelligence, researchers demonstrated a population-based ML model for endometrial cancer. Researchers trained seven ML algorithms based on personal health data from the Prostate, Lung, Colorectal, and Ovarian Cancer Screening Trial (PLCO) and assessed their effectiveness by comparing their performance in stratifying endometrial cancer risk for 100 women with the performance of 15 practicing gynecologic oncologists and primary care physicians.
Among all models, the RF model displayed the best performance by achieving a testing area under the curve (AUC) of 0.96 and was 2.5 times better at identifying above-average risk women with a two-fold reduction in the false-positive rate, indicating a better performance compared to 15 physicians.
In another study published in the journal Frontiers in Artificial Intelligence, researchers demonstrated a novel DL framework for pancreas stereotactic body radiation therapy (SBRT) planning based on 15 test cases and 85 training cases. The framework can predict a fluence map for every beam, avoiding the lengthy inverse optimization process, leading to a 7.1 s average time for fluence-map prediction per patient.
References and Further Reading
Hart, G. R., Yan, V., Huang, G. S., Liang, Y., Nartowt, B. J., Muhammad, W., Deng, J. (2020). Population-Based Screening for Endometrial Cancer: Human vs. Machine Intelligence. Frontiers in Artificial Intelligence, 3, 539879. https://doi.org/10.3389/frai.2020.539879
Wang, W., Sheng, Y., Wang, C., Zhang, J., Li, X., Palta, M., Czito, B., Willett, C. G., Wu, Q., Ge, Y., Yin, F., Wu, Q. J. (2020). Fluence Map Prediction Using Deep Learning Models – Direct Plan Generation for Pancreas Stereotactic Body Radiation Therapy. Frontiers in Artificial Intelligence, 3, 555388. https://doi.org/10.3389/frai.2020.00068
Quazi, S. (2022). Artificial intelligence and machine learning in precision and genomic medicine. Medical Oncology, 39, 120. https://doi.org/10.1007/s12032-022-01711-1
Johnson, K. B., Wei, Q., Weeraratne, D., Frisse, M. E., Misulis, K., Rhee, K., Zhao, J., Snowdon, J. L. (2020). Precision Medicine, AI, and the Future of Personalized Health Care. Clinical and Translational Science, 14(1), 86-93. https://doi.org/10.1111/cts.12884
Bates, A. Leveraging AI for Precision Medicine – Unlocking the Future of Healthcare. [Online] Available at https://eularis.com/leveraging-ai-for-precision-medicine-unlocking-the-future-of-healthcare/ (Accessed on 15 October 2023)