<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Artid Boonrerng on Medium]]></title>
        <description><![CDATA[Stories by Artid Boonrerng on Medium]]></description>
        <link>https://medium.com/@artid.bcu.uk?source=rss-c9b12a363ca3------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/0*jraqLlGrZFfXgRN6</url>
            <title>Stories by Artid Boonrerng on Medium</title>
            <link>https://medium.com/@artid.bcu.uk?source=rss-c9b12a363ca3------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Thu, 19 Nov 2026 16:06:34 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@artid.bcu.uk/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[The Hidden Challenge of Medical Terminology in Healthcare AI]]></title>
            <link>https://medium.com/@artid.bcu.uk/the-hidden-challenge-of-medical-terminology-in-healthcare-ai-e8dff93fc980?source=rss-c9b12a363ca3------2</link>
            <guid isPermaLink="false">https://medium.com/p/e8dff93fc980</guid>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[knowledge-graph]]></category>
            <category><![CDATA[snomed-ct]]></category>
            <category><![CDATA[medical-informatics]]></category>
            <category><![CDATA[healthcare]]></category>
            <dc:creator><![CDATA[Artid Boonrerng]]></dc:creator>
            <pubDate>Tue, 09 Jun 2026 14:34:39 GMT</pubDate>
            <atom:updated>2026-06-09T14:34:39.581Z</atom:updated>
            <content:encoded><![CDATA[<h4>Part 2 of Building Explainable AI for Thai Healthcare</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*38vsXgYX6jWjzuCU_E9nXA.png" /></figure><p>After overcoming the challenge of language, I assumed the next step would be straightforward.</p><blockquote>I was wrong.</blockquote><p>Even when an AI system can correctly identify words in a sentence, it still faces another problem:</p><blockquote>Understanding what those words actually mean.</blockquote><p><strong>In healthcare</strong>, language is only part of the story.</p><p>Medical professionals routinely use abbreviations, synonyms, acronyms, and specialised terminology. Different terms may refer to the same clinical concept, while similar-looking terms can have entirely different meanings.</p><p><strong>For humans</strong>, these differences are often easy to navigate because they are supported by years of education, training, and clinical experience.</p><p><strong>For AI systems</strong>, however, these variations can be surprisingly difficult to interpret.</p><p><strong>More Than Just Words</strong></p><p>Consider the following examples:</p><ul><li>Heart attack</li><li>Myocardial infarction</li><li>MI</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*K3ZXEi0T5ACttORMM11Mcg.png" /></figure><p>For most clinicians, these terms describe the same underlying condition.</p><p>For a machine, they may initially appear as three unrelated pieces of text.</p><p>The problem becomes even more complicated when multiple languages are involved.</p><p><strong>In Thai healthcare</strong>, a single concept may be represented by:</p><ul><li>An English clinical term</li><li>A Thai translation</li><li>A hospital-specific abbreviation</li><li>A coding description used for reimbursement</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*GDmy9j3Pbhqynxbwy5f8Sw.png" /></figure><p>All of these expressions may refer to the same clinical reality.</p><p>Without a way to connect them, AI systems risk treating them as separate concepts.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*NVZsohAYa9GOhQQoOsBdVg.png" /></figure><p><strong>Why Terminology Matters</strong></p><p>This challenge taught me an important lesson.</p><p><strong>Healthcare AI</strong> is not only about processing language.</p><blockquote>It is also about understanding concepts.</blockquote><p>A model may learn patterns from text, but healthcare decisions depend on concepts that remain consistent across different words, documents, and systems.</p><p>This is one reason why healthcare organisations around the world rely on standardised clinical terminologies.</p><p>These standards provide a shared language that allows healthcare professionals, information systems, and AI applications to communicate consistently.</p><p><strong>My First Encounter with SNOMED CT</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*2WojH8U8qBodTtg4qXHQGw.png" /></figure><p>During my PhD, this challenge led me to <strong>SNOMED CT</strong>.</p><p>At first, I viewed it simply as another healthcare terminology system.</p><p>The more I explored it, however, the more I realised that it represented something much larger.</p><blockquote>SNOMED CT was not merely a dictionary of medical terms.</blockquote><p>It was a structured representation of clinical knowledge.</p><p>Concepts were connected to other concepts through meaningful relationships, creating a foundation for semantic understanding rather than simple keyword matching.</p><p>For the first time, I began to see a path beyond traditional text processing.</p><p><strong>Looking Ahead</strong></p><p>Understanding language is important.</p><p>Understanding concepts is even more important.</p><p>My exploration of medical terminology eventually became the foundation for a much deeper question:</p><blockquote>How can we connect clinical concepts in a way that AI systems can understand, explain, and reason about?</blockquote><p>That question would eventually lead me to knowledge graphs.</p><p>In the next article, I will share how SNOMED CT changed the way I think about healthcare knowledge and why terminology systems are far more powerful than most people realise.</p><p>Next in this series:</p><p>Part 3: Building a Thai Healthcare Dataset from Scratch</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=e8dff93fc980" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Why Building AI for Thai Healthcare Is Harder Than You Think]]></title>
            <link>https://medium.com/@artid.bcu.uk/why-building-ai-for-thai-healthcare-is-harder-than-you-think-5fcc61f63838?source=rss-c9b12a363ca3------2</link>
            <guid isPermaLink="false">https://medium.com/p/5fcc61f63838</guid>
            <category><![CDATA[medical-informatics]]></category>
            <category><![CDATA[explainable-ai]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[healthcare]]></category>
            <category><![CDATA[knowledge-graph]]></category>
            <dc:creator><![CDATA[Artid Boonrerng]]></dc:creator>
            <pubDate>Sat, 06 Jun 2026 12:17:13 GMT</pubDate>
            <atom:updated>2026-06-06T12:17:13.482Z</atom:updated>
            <content:encoded><![CDATA[<h4>Part 1 of Building Explainable AI for Thai Healthcare</h4><p>When I began my PhD, I thought the biggest challenge in building AI for Thai healthcare would be the model.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/851/1*j-Eswb1Lzorcs8H-GxQoAw.png" /></figure><blockquote>Better algorithms. More data. Larger language models.</blockquote><p>Like many researchers entering the field of Artificial Intelligence, I believed that technological progress would eventually solve most of the difficult problems. If modern language models could understand English medical texts, surely they could be adapted to Thai healthcare as well.</p><p>Three years later, I realised I had been looking in the wrong direction.</p><p>The hardest part was not the model.</p><blockquote>It was the knowledge.</blockquote><p><strong>The Promise of AI in Healthcare</strong></p><p><strong>Healthcare </strong>is one of the most promising areas for the application of AI.</p><p>From predicting diseases and supporting clinical decisions to automating medical coding, AI has the potential to reduce workload, improve efficiency, and help healthcare professionals make better-informed decisions.</p><p>Recent advances in <strong>Large Language Models (LLMs) </strong>have made this future seem closer than ever. These models can generate medical summaries, answer clinical questions, and retrieve relevant information from vast collections of documents with remarkable fluency.</p><p>It is easy to see why many people are excited about the future of AI in healthcare.</p><p>However, applying these technologies to Thai healthcare is not as straightforward as it might appear.</p><p>Many of the assumptions behind modern AI systems are built on English-language data and resources. When these systems are applied to Thai healthcare, a different set of challenges begins to emerge. These challenges are often invisible to researchers working primarily with English medical texts.</p><p><strong>Challenge 1: Thai Is Not English</strong></p><p>Many <strong>Natural Language Processing (NLP)</strong> techniques were originally developed for languages that use spaces to separate words.</p><blockquote>Thai does not.</blockquote><p>A single sentence can contain multiple medical concepts without clear word boundaries. This means that word segmentation is not just a linguistic challenge; it can directly influence how an AI system understands a patient’s condition and the clinical concepts described in the text.</p><p>For instance, a clinical phrase may contain a disease name, an anatomical structure, and a clinical modifier within a continuous sequence of characters. A seemingly minor tokenisation error can completely alter the intended meaning.</p><p>This becomes especially challenging when the objective is not merely to process text, but to understand a patient’s clinical condition. In such cases, a mistake in word segmentation is not just a language problem; it can affect how an AI system interprets the information it is given.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*KykxSbU0cerTTC-6zMJPlQ.png" /></figure><p><strong>Challenge 2: Medical Language Is a Language of Its Own</strong></p><p>Even after addressing the language problem, another challenge emerges.</p><p>Medical terminology is highly specialised. Healthcare professionals frequently use abbreviations, synonyms, and domain-specific expressions that may refer to the same clinical concept.</p><p>For humans, understanding these variations is often straightforward because they can draw upon contextual and domain knowledge. For AI systems, however, these expressions may appear to represent entirely different concepts.</p><p>This is why healthcare systems around the world rely on standardised terminologies and data standards, including:</p><ul><li><strong>ICD-10</strong> for disease classification,</li><li><strong>SNOMED CT</strong> for representing clinical concepts, and</li><li><strong>HL7 FHIR</strong> for healthcare data exchange and interoperability.</li></ul><p>Without linking language to these structured knowledge systems, AI risks becoming little more than an advanced text generator rather than a reliable tool for supporting healthcare decision-making.</p><p><strong>Challenge 3: Accuracy Is Not Enough</strong></p><p>One lesson from my research surprised me more than anything else.</p><p>In healthcare, being correct is often <em>not enough</em>.</p><p>Clinicians, coders, auditors, and healthcare organisations need to understand why a recommendation was made.</p><p>A model that suggests a diagnosis code without providing any supporting evidence may achieve high accuracy, but it can still be difficult to trust in real-world clinical settings.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*f7YwrDU4ZMW_w0BwVEWOLw.png" /></figure><p>This realisation led me towards a different research direction.</p><p>Instead of asking:</p><blockquote>Can AI produce the correct answer?</blockquote><p>I began asking:</p><blockquote>Can AI explain why it produced that answer?</blockquote><p><strong>From Language Models to Knowledge Graphs</strong></p><p>This question eventually led me to explore <strong>Knowledge Graphs</strong>.</p><p>At its core, a Knowledge Graph represents concepts and the relationships between them. Rather than storing information as isolated pieces of text, it organises knowledge into connected structures that can be inspected, validated, and traced.</p><p>What interested me most was the potential for transparency. In healthcare, it is often important to understand why a recommendation was made, not just what the recommendation is.</p><p>The goal is not to replace language models. Instead, it is to combine the strengths of language models with structured medical knowledge, allowing AI systems to provide answers that are not only useful, but also easier to understand and trust.</p><p><strong>Looking Ahead</strong></p><p>My PhD journey began with Thai medical text.</p><p>Over time, it grew into a broader exploration of medical terminologies, clinical coding, semantic interoperability, and explainable AI.</p><p>Along the way, I realised that building AI for healthcare is not simply a machine learning challenge.</p><blockquote>It is a knowledge challenge.</blockquote><p>In this series, I will share lessons from building datasets, mapping Thai healthcare concepts to SNOMED CT, developing healthcare knowledge graphs, and exploring how explainable AI can help create more trustworthy healthcare systems.</p><blockquote>The journey is still ongoing.</blockquote><p>But it has already changed the way I think about what AI should be.</p><p>Next in this series:</p><p>Part 2: The Hidden Challenge of Thai Medical Language</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=5fcc61f63838" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>