Medalyze: Lightweight Medical Report Summarization Application Using FLAN-T5-Large
VAN-TINH NGUYEN 1, HOANG-DUONG PHAM 2, THANH-HAI TO 2, CONG-TUAN HUNG DO 2, THI-THU TRANG DONG 3, VU-TRUNG DUONG LE 4, VAN-PHUC HOANG 1 · 2025
Read Medalyze: Lightweight Medical Report Summarization Application Using FLAN-T5-Large by VAN-TINH NGUYEN 1, HOANG-DUONG PHAM 2, THANH-HAI TO 2, CONG-TUAN HUNG DO 2, THI-THU TRANG DONG 3, VU-TRUNG DUONG LE 4, VAN-PHUC HOANG 1 (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2505-17059.txt.
Introduction
Medical reports are written for doctors, not patients, and their dense wording leaves many readers lost. Van-Tinh Nguyen and co-authors built Medalyze, a lightweight app that puts plain-language help on web and mobile. Three fine-tuned FLAN-T5-Large models do the work: one summarizes reports, one pulls health issues out of patient-doctor conversations, and one finds the key question inside a passage. The system answers in real time through a scalable API backed by a distributed database.
In tests, Medalyze beat GPT-4 on domain-specific summarization across standard scoring metrics. The authors pitch it as a private, practical way to make health information readable without sending data to giant general models.
Catalog record
| CREATOR | VAN-TINH NGUYEN 1, HOANG-DUONG PHAM 2, THANH-HAI TO 2, CONG-TUAN HUNG DO 2, THI-THU TRANG DONG 3, VU-TRUNG DUONG LE 4, VAN-PHUC HOANG 1 |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 57 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2505-17059
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Medalyze: Lightweight Medical Report Summarization Application Using FLAN-T5-Large VAN-TINH NGUYEN 1, HOANG-DUONG PHAM 2, THANH-HAI TO 2, CONG-TUAN HUNG DO 2, THI-THU TRANG DONG 3, VU-TRUNG DUONG LE 4, VAN-PHUC HOANG 1 Abstract Understanding medical texts presents significant challenges for both healthcare professionals and patients due to the complexity of terminology and the context-dependent nature of medical language. To address this issue, this paper introduces Medalyze, an AI-powered application designed to enhance the comprehension of medical texts through advanced natural language processing (NLP) techniques. Medalyze leverages three specialized Flan-T5-Large models, each fine-tuned for a distinct task: (1) summarizing key information from medical reports, (2) extracting health-related issues from conversational exchanges, and (3) identifying the primary question within a given medical text. The application offers a multi-platform solution, featuring both a web-based interface and a mobile application, ensuring accessibility for a diverse user base. Data management is powered by YugabyteDB, enabling seamless synchronization across platforms. To ensure efficient performance, Medalyze is deployed using a scalable API architecture, providing real-time text processing. A comprehensive evaluation was conducted to assess the performance of Medalyze against GPT-4, demonstrating superior performance in specialized medical text summarization tasks. User feedback indicates significant improvements in understanding medical passages, with enhanced clarity and usability. Medalyze represents a practical and efficient tool for improving communication and decision-making in healthcare settings, bridging the gap between complex medical language and user-friendly information for both professionals and non-specialist audiences. † † history: Date of publication xxxx 00, 0000, date of current version xxxx 00, 0000. † † doi: 10.1109/ACCESS.2024.0429000 † † address: Electrical Engineering Department, Le Quy Don Technical University, Ha Noi, Viet Nam ( [email protected]) † † address: Department of Information Technology, University of Science and Technology of Hanoi, Ha Noi, Viet Nam † † address: Department of acute diseases and Emergency - Institute for the treatment of senior Staff, 108 Institute of Clinical Medical and Pharmaceutical Sciences, Ha Noi, Viet Nam † † address: Computing Architecture Laboratory, Nara Institute of Science and Technology, Nara, Japan † † corresponding: Corresponding author: Van-Phuc Hoang ([email protected]). I Introduction Medical transcriptions are essential for healthcare, providing detailed information about patient conditions, diagnostic results, and treatment plans [ 1 ] – [ 3 ] . However, the inherent complexity of medical terminology and the extensive nature of these documents present significant challenges [ 4 ] – [ 6 ] . Medical professionals often face difficulties in efficiently identifying key insights within lengthy reports, while patients and caregivers may struggle to understand specialized language and technical details, hindering informed decision-making and proactive health management. The rapid digitization of healthcare records, coupled with the increasing adoption of AI-driven tools, underscores the need for solutions that enhance the comprehension of medical content [ 7 ] – [ 15 ] . As a response to this demand, Medalyze is introduced as an AI-powered application designed to bridge the communication gap between healthcare providers and non-specialist audiences. By leveraging cutting-edge natural language processing (NLP) techniques, the application facilitates the interpretation of medical texts without compromising accuracy, thereby enhancing accessibility for a diverse range of users. Medalyze not only aids healthcare professionals by reducing the cognitive load associated with interpreting medical transcriptions but also empowers patients by simplifying complex information. This paper provides a comprehensive analysis of the technical development, key features, and performance evaluation of Medalyze, highlighting its potential to transform the way medical content is accessed and understood. By addressing key problems in healthcare communication, Medalyze aims to improve decision-making, enhance patient outcomes, and foster a more inclusive healthcare ecosystem. The main contributions of this study are given as: • Introducing Medalyze, an AI-powered application leveraging three specialized Flan-T5-Large models, each tailored to perform distinct tasks: 1. Summarizing key information in medical reports (text format); 2. Extracting health issues from conversational exchanges; and 3. Identifying the central question within a given passage. • Establishing a multi-platform ecosystem for Medalyze, featuring a web-based interface (developed using HTML, CSS, and JavaScript) and a mobile application for Android (built using Java). The underlying data management is powered by YugabyteDB, enabling efficient data synchronization and sharing across platforms. APIs ensure smooth communication between system components, enhancing the usability and reliability of the application. • Conducting a performance analysis of Medalyze within the healthcare ecosystem, with a comparative evaluation against GPT-4. The remainder of this paper is organized as follows: The related works are presented in Section II. The background and research methodology used to build Medalyze are presented in Sections III and IV. Next, Sections V and VI cover the source of dataset acquisition, data processing, and pre-trained models. Section VII introduces the evaluation metrics, while Section VIII details the web and mobile application architecture. Section IX provides a comparative analysis with GPT-4. Finally, Section X concludes the paper. II Related works The use of AI and NLP in processing medical texts has seen significant growth in recent years, with numerous projects and research initiatives aiming to enhance healthcare communication [ 16 ] – [ 22 ] . One prominent example is the application of advanced AI models like ChatGPT, Claude, and Gemini, which have been explored for their ability to alleviate clinical documentation burdens by processing and summarizing complex clinical texts [ 23 ] – [ 35 ] . These models demonstrate the potential of AI to streamline workflows and improve the accessibility of medical content. Another significant advancement involves the adoption of FLAN-T5 models, which have been applied to tasks such as transforming clinical text into structured Fast Healthcare Interoperability Resources (FHIR) formats and summarizing conversations to enhance information accessibility [ 36 ] – [ 38 ] . These models highlight the versatility of fine-tuned language models in addressing a wide range of medical text processing requirements. Efforts by organizations like Dataloop have resulted in the development of specialized AI tools capable of summarizing complex medical documents, research papers, and clinical notes into concise, comprehensible formats. These solutions utilize pre-trained transformer models and extensive medical literature to enhance their efficacy, making them valuable assets in the medical field. For instance, a study conducted at Stanford University [ 38 ] investigated the use of large language models (LLMs) for summarizing clinical text. Notably, the use of Flan-T5-XL in this study demonstrated competitive performance despite having a relatively low parameter count of 2.7 billion, compared to models like Alpaca, Vicuna, Llama-2, and even GPT-4. This competitive performance is attributed to the model’s architecture, specifically its Sequence-to-Sequence (Seq2Seq) design, which is inherently well-suited for text summarization tasks. This result was a key motivation for selecting Flan-T5-Large in Medalyze, ensuring a lightweight application with quantifiable accuracy and contextual consistency. Public reception of these technologies has been largely positive, with studies showing that adapted LLMs can outperform humans in certain summarization tasks [ 39 ] – [ 50 ] . This demonstrates their potential to significantly enhance efficiency in clinical workflows, reduce cognitive burdens on medical professionals, and empower patients with clearer, more actionable medical information. However, challenges such as ensuring data privacy, maintaining accuracy, and addressing ethical concerns remain central to the ongoing development and adoption of these tools. Medalyze builds upon these advancements, distinguishing itself by integrating three specialized AI models for distinct tasks: