Browsing by Subject "Boolean Logic Synthesis"

Now showing 1 - 1 of 1
  • Thumbnail Image
    Item
    Towards a transparent OmniDoctor: AI assistant for clinical decision support
    (2024-01) Sahoo, Himanshu Shekhar
    This thesis presents OmniDoctor, an advanced interpretable Artificial Intelligence (AI) assistant, engineered to revolutionize clinical decision support (CDS) in healthcare. At its core, OmniDoctor integrates diverse data types - including textual information, and structured data from Electronic Health Records (EHRs) - to enhance patient care, demonstrating notable advancement in the application of AI in real-world clinical scenarios. OmniDoctor distinguishes itself with its commitment to transparency and interpretability, key components in modern healthcare AI. This approach aligns with the growing demand for AI tools that augment human expertise, ensuring that AI-generated recommendations are both accurate and understandable. This is particularly crucial in healthcare, where decision-making can have profound implications. During the COVID-19 pandemic, the effectiveness and versatility of OmniDoctor is brought into limelight. OmniDoctor adeptly utilizes Natural Language Processing (NLP) to navigate the complexities of COVID-19 symptomatology, providing robust and timely support for clinical decision-making. It transcends traditional symptom extraction methods by employing advanced NLP techniques to analyze unstructured clinical notes, thereby synthesizing a comprehensive feature set that is responsive to the dynamic and evolving nature of the pandemic. In addition to this, OmniDoctor integrates lightweight, pre-trained transformer models to enhance the efficiency and scalability of COVID-19 symptom identification. This integration marks a significant shift from the conventional, resource-intensive NLP systems, offering a more rapid and adaptable response to the pandemic’s challenges. Simultaneously, in its adaptive learning phase, OmniDoctor employs a fine-tuned Bidirectional Encoder Representations from Transformers (BERT) model for Named Entity Recognition (NER). This reflects the system’s capacity for ongoing learning and adaptability, crucial traits in medical practice. Such an approach enables OmniDoctor to stay abreast with the mutating virus, ensuring precise symptom annotation and continuous evolution to meet the demands of the healthcarelandscape during the pandemic. To realize the goal of an advanced AI assistant, this research expands the scope of OmniDoctor to include critical conditions such as acute coronary syndrome (ACS), showcasing its application far beyond infectious diseases like COVID-19. In this realm, OmniDoctor concentrates its effort towards achieving diagnostic equity and ethical responsibility for potential life-threatening diagnoses like ACS. Its focus on interpretability and fairness is instrumental in delivering AI-driven insights facilitating equitable and well-informed patient care. To further enhance the diagnostic capabilities of OmniDoctor, this research further expands to complex chest pain diagnoses (CPDs), where OmniDoctor is implemented to distinguish between 13 different conditions, each presenting unique diagnostic challenges. It utilizes boolean function ensembles (BFEs) to transform complex AI analyses into insights accessible to clinicians. This advancement is particularly vital in emergency care settings, where interpretability of the decisions made by OmniDoctor can potentially allow for rapid and accurate diagnoses, significantly improving clinical decision-making processes. OmniDoctor’s journey towards achieving its full potential is demonstrated in its approach to handling multiple critical chest pain conditions through the utilization of unstructured clinical notes. The application of pre-trained Large Language Models (LLMs), fine-tuned on clinical notes, marks a notable stride in OmniDoctor’s evolution. This advanced approach enables OmniDoctor not only to deliver precise diagnostics but also to provide rich, comprehensive explanatory narratives. These narratives effectively resonate with the thought processes of Emergency Department (ED) experts, thereby enhancing theinterpretability and relevance of OmniDoctor’s insights. In summary, OmniDoctor stands as a groundbreaking tool in AI-driven healthcare, evolving from a predictive assistant to an interpretative partner for healthcare providers. We envision OmniDoctor to be a healthcare provider’s personal assistant, seamlessly integrating into their workflow and offering vital support in decision-making processes. By combining advanced AI technologies with a deep understanding of clinical needs, OmniDoctor not only fosters trust in AI applications in healthcare but also sets a new standard in patient-centered, AI-enhanced medical care. With OmniDoctor, this research strives to contribute to more personalized, efficient, and effective healthcare delivery, where technology and humanexpertise work in unison for the betterment of patient outcomes.