Applications of AI in Healthcare
By Grace Liberty
Abstract:
In the ever-evolving landscape of healthcare, the significance of cybersecurity cannot be overstated, as it plays a pivotal role in both preserving the integrity of services and safeguarding sensitive patient data. With the continuous digitization of healthcare, including the widespread adoption of electronic health records and the integration of new medical devices, the need for robust protection against emerging cyber threats becomes increasingly urgent. This urgency is further emphasized by the integration of artificial intelligence (AI) as a crucial tool in healthcare diagnostics, clinical practice, clinical data management, and further automation and enhanced need for cybersecurity. This blog post aims to explore the multifaceted application of AI in healthcare. The overarching objective is to explore how AI technologies contribute to the enhancement of contemporary healthcare practices, how they impact cybersecurity, and how they provide insights into the promising and concerning future of AI in healthcare.
Introduction
The convergence of artificial intelligence (AI), Federated Learning (FL), and healthcare creates a blend reshaping the landscape of cybersecurity, diagnostics, and health monitoring. In an era marked by the rapid digitization of healthcare, the protection of patient data and the fortification of healthcare services against evolving cyber threats have become critical imperatives. AI is a versatile tool. It has been integrated into healthcare systems and biomedical technologies, with further integration planned in the near future. Its impact goes beyond data management, extending into the domain of healthcare diagnostics and health management. This integration signifies a shift poised to revolutionize the industry by augmenting the speed and precision of disease detection, prognosis, and treatment recommendations. Despite its transformative potential, there are concerns about data access, privacy, interpretability, and biases. Notably, AI-driven diagnostic tools are revolutionizing the reading of CT and MRI scans, enabling faster analyses and automating notetaking. AI-driven diagnostic tools, equipped to analyze expansive datasets and quickly discern patterns and anomalies in medical imaging, clinical data, and patient histories, promise to usher in a new era of more precise and timely patient care. However, caution is necessary, considering potential biases and the need for continuous validation in AI integration for healthcare.
Federated Learning is a machine learning approach that allows a model to be trained across multiple decentralized devices or servers holding local data samples without exchanging them. It introduces a collaborative and privacy-focused framework, addressing the need to safeguard sensitive medical information during data exchange. However, challenges emerge, such as the reliability of decentralized models and logistical hurdles in accessing diverse data for training. The journey towards technologically advanced healthcare practices requires a balanced and comprehensive approach to address both promises and challenges of AI, FL, and patient care. This blogpost will proceed by briefly addressing AI/Federated Learning cybersecurity, diagnostics, and personalized healthcare.
AI/Federated Learning Cybersecurity
Watermarking is a technique utilized by AI and allows for stamping a document to prove its authenticity before diagnosis. Watermarking was initially applied for broadcast monitoring and proof of ownership. It has since evolved into a vital tool for preserving patient privacy, document authenticity, and enhanced cybersecurity. Steganography is another technique utilized to conceal encrypted messages and applied in healthcare to safeguard clinical data.[i] By embedding unique watermarks into machine learning models, federated learning ensures collaborative training without centralizing sensitive patient information.[ii] Watermarking serves as a privacy-preserving mechanism, maintaining confidentiality, reliability, and availability of medical images.[iii] The demands imposed by ethical and legislative standards in healthcare cybersecurity accentuate the need for sophisticated watermarking and steganography techniques. The integration of these techniques within federated learning models showcases its pivotal role in advancing secure and collaborative medical research.
The decentralized model in the intersection of cybersecurity and machine learning (ML) is gaining significant attention, particularly regarding the Internet of Things (IoT).[iv] The concept of Federated Cybersecurity (FC) promises to enhance the safety and efficiency of IoT systems and medical IoT devices. This approach aims to efficiently detect security threats, initiate countermeasures, and curtail the propagation of threats throughout the IoT network by forming federations of learned and shared models from various participants.[v] Centralized learning and Federated Learning emphasize the role of FL as a privacy-aware ML model crucial for securing the vulnerable IoT environment.[vi] It focuses primarily on the security aspect, addressing various approaches that tackle performance issues such as accuracy, latency, and resource constraints associated with FL. These performance considerations are essential as they can impact both the security and overall efficiency of the IoT. However, the considerable data requirements and time-intensive nature of training models for specific tasks within the healthcare and IoT domains pose challenges. Addressing these challenges is vital to ensuring the effectiveness and efficiency of federated learning in real-world applications. FL responds to different cyberattacks by employing strong security measures, including encryption and continuous monitoring, to safeguard decentralized model updates and ensure the privacy of sensitive medical data.[vii] This knowledge leads us to anticipate the future evolution of this decentralized paradigm and navigate the intricate landscape of securing IoT ecosystems.
Diagnostics
The expansion of AI in healthcare, especially its diagnostic capabilities, is transformative as it brings significant advancements and creates discussions about the future potential of further AI integration along with related concerns and issues. AI use in healthcare is a paradigm shift.[viii] It is seen by many as a crucial step in addressing healthcare disparities in rural and underserved areas as well as among racial minorities.[ix] For example, the approach to Diabetic Retinopathy (DR) detection is undergoing a transformation through the incorporation of Federated Learning in the DRFL (Diabetic Retinopathy Federated Learning) model.
Diabetic Retinopathy, a complication arising from diabetes, leads to retinal lesions and demands early identification to prevent irreversible vision loss.[x] The DRFL model uses the collaborative potential of FL, allowing deep learning (DL) models to train across various medical institutions with a specific emphasis on utilizing AI to interpret medical scans. This model strategically combines Federated Averaging (FedAvg) with the median categorical cross-entropy loss, ensuring efficient grading of DR severity and prioritizing the privacy of patient data.[xi] The innovative central server of DRFL excels in extracting multi-scale features from fundus images, resulting in an impressive accuracy rate of 98.6%, specificity of 99.3%, precision of 97.25%, and an F1 score of 97.5%—outperforming existing techniques.[xii] Addressing the scarcity of ophthalmologists, DRFL emerged as an automated and collaborative tool, promising a significant leap in the diagnostics of diabetic Retinopathy.
Further, using AI to evaluate cardiovascular health is another opportunity in clinical diagnostics that is gaining traction to detect congestive heart failure (CHF). Heart failure is a pervasive global health challenge that affects three to five individuals per hundred and carries a staggering mortality rate of up to 50% within five years.[xiii] Recognizing the urgency of early detection, especially at the elusive early stages, is paramount in mitigating the progression and improving survival rates.[xiv] Despite its critical importance, accurate diagnosis, particularly in the early stages, is a challenge. The New York Heart Association (NYHA) categorizes heart failure into four levels, with significant symptoms emerging only in patients at levels III and IV.[xv] The intricacies of diagnosing heart failure, compounded by subjectivity and time-intensive processes, necessitate an improvement in analytical methodologies.
Traditionally, Electrocardiogram (ECG) stands as a cornerstone in heart rhythm analysis, offering detailed insights into cardiac activities.[xvi] However, manual decoding of ECG signals proves laborious and introduces a subjective element reliant on the clinician's expertise. In order to address these challenges, machine learning models have been harnessed for automated CHF diagnosis, particularly leveraging long-term ECG signals.[xvii] This shift towards automated diagnosis expedites the diagnostic process, reduces opportunities for human error, and provides for consistent diagnostic methodology. These innovative approaches signify a transformative leap towards enhancing CHF diagnosis, offering hope for enhanced efficiency and improved patient outcomes.
Personalized Healthcare
Another area where artificial intelligence is making an impact and holds a lot of promise is in personalizing medical treatments, changing how healthcare professionals analyze patient data, identifying disease patterns, and tailoring medical interventions. One notable example stands out in AI application in personalized healthcare. In 2015, three entities successfully collaborated (Atomwise corporation, IBM, and the University of Toronto) to apply an AI-powered program to screen the available drugs that could be "re-designed" to combat the West African Ebola virus outbreak.[xviii] The AI-powered program was able to quickly analyze vast amounts of data across compound datasets to identify potential treatments. It evaluated the compounds' interactions with the glycoprotein, which is important for preventing the Ebola virus's penetration into cells.[xix] By assessing vast amounts of data in a short period, AI expedited the identification of a compound with the desired characteristics and was able to help researchers quickly test the compound, produce drugs, and prevent the global pandemic.
In the context of personalized treatments, AI is used to leverage vast datasets, including patient records, genetic information, and molecular data, to identify unique patterns and correlations.[xx] This capability allows AI systems to gain insights into individual patient characteristics, responses to specific treatments, and genetic predispositions, ultimately enabling the customization of medical interventions. Scholars are discussing the potential of AI to advance personalized medicine but also caution about the limitations of many AI techniques.[xxi]
In broader healthcare applications, AI-driven precision medicine involves analyzing diverse data types, such as genomics, proteomics, and patient records, to identify the most effective treatment.[xxii] Machine learning algorithms can recognize subtle patterns and correlations within these datasets, predicting how patients may respond to specific therapies based on their unique biological makeup. AI also has the potential to impact the prevention and early detection of diseases in the human population, which could decrease the disease burden for the public and "the cost of preventable health care for all."[xxiii] Moreover, AI can facilitate real-time monitoring and analysis of patient data collected through Internet of Medical Things (IoMT) devices. This continuous stream of data provides healthcare professionals with insights into patients' health metrics, enabling the timely adjustment of treatment plans to align with individual needs.
However, despite these remarkable advancements, it's crucial to acknowledge the downsides of AI in healthcare. One notable concern is the potential for bias in AI algorithms, as they rely heavily on historical data that may perpetuate existing disparities in healthcare.[xxiv] If not carefully addressed, these biases could result in inaccurate treatment recommendations and exacerbate healthcare disparities. Another challenge lies in the interpretability of AI-driven models. The complex nature of machine learning algorithms makes it difficult for healthcare professionals to fully understand the rationale behind AI-generated recommendations. This lack of transparency may lead to hesitancy in trusting AI systems, hindering their widespread adoption in clinical decision-making.
Conclusion
In conclusion, the integration of Artificial Intelligence (AI) in healthcare and advancements in Federated Learning (FL) has opened a transformative era marked by enhanced cybersecurity, diagnostics, and personalized healthcare. The convergence of these technologies addresses the critical need for safeguarding patient data, fortifying healthcare services against evolving cyber threats, and revolutionizing medical diagnostics. There are also significant hurdles to overcome before AI and FL gain more efficiency, fidelity, and trust. As we navigate this new era of technological opportunities, both good and bad, the potential for further innovations in healthcare remains vast, promising a future where advanced technologies contribute significantly to improved patient care, improved health outcomes, and reduced overall costs of healthcare.
[i] Mousavi, S. M., Naghsh, A., & Abu-Bakar, S. A. (2014). Watermarking techniques used in medical images: a survey. Journal of digital imaging, 27(6), 714–729. https://doi.org/10.1007/s10278-014-9700-5.
[ii] Zain, J., & Clarke, M. (2020). Security In Telemedicine: Issues In Watermarking Medical Images. Sciences of Electronic, Technologies of Information and Telecommunications. https://www.readcube.com/library/b0cb7c25-631b-4713-81f8-3bb909eb9dcc:0346abb7-4fd3-4cc5-8a15-8e58f6991b79.
[iii] P. Singh, K. J. Devi, H. K. Thakkar and K. Kotecha, "Region-Based Hybrid Medical Image Watermarking Scheme for Robust and Secured Transmission in IoMT," in IEEE Access, vol. 10, pp. 8974-8993, 2022, doi: 10.1109/ACCESS.2022.3143801.
[iv] AlZubi, A.A., Al-Maitah, M. & Alarifi, A. Cyber-attack detection in healthcare using cyber-physical system and machine learning techniques. Soft Comput 25, 12319–12332 (2021). https://doi.org/10.1007/s00500-021-05926-8.
[v] H. Elayan, M. Aloqaily and M. Guizani, "Sustainability of Healthcare Data Analysis IoT-Based Systems Using Deep Federated Learning," in IEEE Internet of Things Journal, vol. 9, no. 10, pp. 7338-7346, 15 May 15, 2022, doi: 10.1109/JIOT.2021.3103635.
[vi] Z. E. Ekolle, R. Kohno and H. Ochiai, "A Distributed Cybersecurity Solution in an IoMT Network Using a Multi-target Federated Learning," 2023 IEEE 17th International Symposium on Medical Information and Communication Technology (ISMICT), Lincoln, NE, USA, 2023, pp. 1-6, doi: 10.1109/ISMICT58261.2023.10152107.
[vii] B. Ghimire and D. B. Rawat, "Recent Advances on Federated Learning for Cybersecurity and Cybersecurity for Federated Learning for Internet of Things," in IEEE Internet of Things Journal, vol. 9, no. 11, pp. 8229-8249, 1 June 1, 2022, doi: 10.1109/JIOT.2022.3150363.
[viii] Kuhn, Thomas S., 1922-1996, author. The Structure of Scientific Revolutions. Chicago :University of Chicago Press, 1970.
[ix] d'Elia, A., Gabbay, M., Rodgers, S., Kierans, C., Jones, E., Durrani, I., Thomas, A., & Frith, L. (2022). Artificial intelligence and health inequities in primary care: a systematic scoping review and framework. Family medicine and community health, 10(Suppl 1), e001670. https://doi.org/10.1136/fmch-2022-001670.
[x] Lo, J., Yu, T. T., Ma, D., Zang, P., Owen, J. P., Zhang, Q., Wang, R. K., Beg, M. F., Lee, A. Y., Jia, Y., & Sarunic, M. V. (2021). Federated Learning for Microvasculature Segmentation and Diabetic Retinopathy Classification of OCT Data. Ophthalmology science, 1(4), 100069. https://doi.org/10.1016/j.xops.2021.100069
[xi] N. J. Mohan, R. Murugan, T. Goel and P. Roy, "DRFL: Federated Learning in Diabetic Retinopathy Grading Using Fundus Images," in IEEE Transactions on Parallel and Distributed Systems, vol. 34, no. 6, pp. 1789-1801, June 2023, doi: 10.1109/TPDS.2023.3264473.
[xii] N. J. Mohan, R. Murugan, T. Goel and P. Roy, "DRFL: Federated Learning in Diabetic Retinopathy Grading Using Fundus Images," in IEEE Transactions on Parallel and Distributed Systems, vol. 34, no. 6, pp. 1789-1801, June 2023, doi: 10.1109/TPDS.2023.3264473.
[xiii] L. Zou, Z. Huang, X. Yu, J. Zheng, A. Liu and M. Lei, "Automatic Detection of Congestive Heart Failure Based on Multiscale Residual UNet++: From Centralized Learning to Federated Learning," in IEEE Transactions on Instrumentation and Measurement, vol. 72, pp. 1-13, 2023, Art no. 4001013, doi: 10.1109/TIM.2022.3227955.
[xiv] D.-J. Choi, J. J. Park, T. Ali, and S. Lee, “Artificial intelligence for the diagnosis of heart failure,” NPJ Digit. Med., vol. 3, no. 1, p. 54, Apr. 2020.
[xv] B. Bozkurt et al., “Universal definition and classification of heart failure,” J. Cardiac Failure, vol. 27, no. 4, pp. 387–413, 2021.
[xvi] J. Zhang et al., “MLBF-Net: A multi-lead-branch fusion network for multi-class arrhythmia classification using 12-lead ECG,” IEEE J. Transl. Eng. Health Med., vol. 9, 2021, Art. no. 1900211.
[xvii] V. Jahmunah et al., “Computer-aided diagnosis of congestive heart failure using ECG signals—A review,” Phys. Medica, vol. 62, pp. 95–104, Jun. 2019.
[xviii] Kwo, L. (2021, July 2). Contributed: Top 10 use cases for AI in Healthcare. MobiHealthNews. https://www.mobihealthnews.com/news/contributed-top-10-use-cases-ai-healthcare.
[xix] Kumar, A., Gadag, S., & Nayak, U. Y. (2021). The Beginning of a New Era: Artificial Intelligence in Healthcare. Advanced pharmaceutical bulletin, 11(3), 414–425. https://doi.org/10.34172/apb.2021.049.
[xx] D. Gupta, M. P. S. Bhatia and A. Kumar, "Resolving Data Overload and Latency Issues in Multivariate Time-Series IoMT Data for Mental Health Monitoring," in IEEE Sensors Journal, vol. 21, no. 22, pp. 25421-25428, 15 Nov.15, 2021, doi: 10.1109/JSEN.2021.3095853.
[xxi] Schork N. J. (2019). Artificial Intelligence and Personalized Medicine. Cancer treatment and research, 178, 265–283. https://doi.org/10.1007/978-3-030-16391-4_11.
[xxii] Johnson, K. B., Wei, W. Q., Weeraratne, D., Frisse, M. E., Misulis, K., Rhee, K., Zhao, J., & Snowdon, J. L. (2021). 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
[xxiii] Johnson, K. B., Wei, W. Q., Weeraratne, D., Frisse, M. E., Misulis, K., Rhee, K., Zhao, J., & Snowdon, J. L. (2021). 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.
[xxiv] Celi, L. A., Cellini, J., Charpignon, M. L., Dee, E. C., Dernoncourt, F., Eber, R., Mitchell, W. G., Moukheiber, L., Schirmer, J., Situ, J., Paguio, J., Park, J., Wawira, J. G., Yao, S., & for MIT Critical Data (2022). Sources of bias in artificial intelligence that perpetuate healthcare disparities-A global review. PLOS digital health, 1(3), e0000022. https://doi.org/10.1371/journal.pdig.0000022.