Chérubin Mugisha

Mr. Chérubin Mugisha

orcid

Research Focus & Metrics

Academic Career Length

7Years (2020-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
31 January 2022 in IEEE Access
Verified
Journal Article
Open access
CMChérubin Mugisha
Chérubin Mugisha
Incheon PaikIncheon Paik
Incheon Paik
Machine learning techniques and algorithm-based approaches are becoming more and more vital to support clinical decision-making. In the medical area, natural language processing (NLP) techniques have shown the ability to extract useful information from electronic health records. On the one hand, statistic, semantic, and contextualized word embedding-based models and on the other hand preprocessing approaches are the keys to a better representation of a document. Using narratives from the Intensive Care Unit, we elaborated a comparison of the most used methods and preprocessing approaches to tackle an outcome prediction problem and guide researchers into NLP pipelines in the medical area.We used real data from Medical Information Mart for Intensive Care-III (MIMIC-III).We selected all notes related to patients with pneumonia. We conducted a deep analysis on text preprocessing tasks producing three datasets: raw data with minor preprocessing, meticulous preprocessing, and extreme preprocessing filtering only medical-related terminologies using Named Entity Recognition algorithms. We then used these three sets in five models, of which two are based on the traditional noncontextual word embedding techniques and three use contextualized word embedding based on a transformer. We demonstrated that transformer-based models outperform other word embedding models and a profound preprocessing yielded an accuracy of 98.2 F1-score. These results show the highly competitive ability of NLP predictive models against other models that use medical data. With an appropriate NLP pipeline, the information contained in medical narratives can be used to draw up a patient profile, and admission notes can help to ascertain a mortality risk of a patient admitted to the Intensive Care Unit.

16 May 2024 in IEEE Access
Verified
Journal Article
Open access
HDHong N. Dao
Hong N. Dao
TNTuyen Nguyen
Tuyen Nguyen
CMChérubin Mugisha
Chérubin Mugisha
Incheon PaikIncheon Paik
Incheon Paik
Medical image data often face the problem of data scarcity and costly annotation processes. To overcome this, our study introduces a novel transfer learning method for medical image classification. We present a multimodal learning framework that incorporates the pre-trained PubMedCLIP model and multimodal feature fusion. Prompts of different complexities are combined with images as inputs to the proposed model. Our findings demonstrate that this approach significantly enhances image classification tasks while reducing the burden of annotation costs. Our study underscores the potential of PubMedCLIP in revolutionizing medical image analysis through its prompt-based approach and showcases the value of multi-modality for training robust models in healthcare. Code is available at: https://github.com/HongJapan/MTL_prompt_medical.git .

16 December 2020
Verified
Conference Paper
Closed access
CMChérubin Mugisha
Chérubin Mugisha
IPIncheon Paik
Incheon Paik
In Intensive Care Unit (ICU), it is important to anticipate interventions for patients at a high-risk of death. This requires identifying those patients ideally at the time of their admission in ICU, and update their initial risk rate every time new data is available. This predictive task can be performed by analyzing structured and unstructured routine data to make sure that we can initiate a prediction for every patient. Traditional statistical tools have been used to assess disease like pneumonia and predict the outcome of a patient. Recently, machine learning models emerged and have shown better performances on such tasks. Although authors have published various results, their works rely on a single datatype, either structured or unstructured data. Using the Medical Information Mart for intensive Care data-set, we are proposing an ensemble model, that aggregates different data-types to predict the outcome of a pneumonia patient admitted in ICU using limited data that can be available at the very early stage of his stay. To demonstrate the importance of this approach, we compared it with 2 different models, one based only on structured data, and another one based on narratives text from caregivers, and we were able to show that our ensemble model can perform way better with an accuracy of 0.98 of F1-score(0.97 MCC), while a model using only structured data had 0.79 of F1-score and where text notes predicted the outcome with an accuracy of 0.89 of Matthews Correlation Coefficient. In addition to showing how ensemble learning models can outperform other models on this task, we demonstrated the importance and usefulness of interpreting the predictions pointing out the leading factors that are determining the global and individual outcome predictions.

13 April 2023 in Electronics
Verified
Journal Article
Open access
PDF file
CMChérubin Mugisha
Chérubin Mugisha
Incheon PaikIncheon Paik
Incheon Paik
The increasing use of electronic health records (EHRs) generates a vast amount of data, which can be leveraged for predictive modeling and improving patient outcomes. However, EHR data are typically mixtures of structured and unstructured data, which presents two major challenges. While several studies have focused on using machine learning models to predict patient outcomes, these models often require data to be in a structured format, which may lead to the loss of important information. On the other hand, unstructured data, such as narrative reports, can be noisy and challenging for natural language processing applications and interoperability. Therefore, there is a need to bridge the gap between structured EHR data and NLP-based predictive models. In this paper, we propose a fuzzy-logic-based pipeline that generates medical narratives from structured EHR data and evaluates its performance in predicting patient outcomes. The pipeline includes a feature selection operation and a reasoning and inference function that generates medical narratives. We then extensively evaluate the generated narratives using transformer-based NLP models for a patient-outcome-prediction task. We furthermore assess the interpretability of the generated text using Shapley values. Our approach has demonstrated comparable performance to the benchmark baseline models with an F1-score of 93.7%, while exhibiting slightly improved results in terms of recall. The model demonstrated proficiency in the preservation of information and interpretability inherited from nuanced and structured narratives. To the best of our knowledge, this is the first study to demonstrate the ability to transform tabular data into text to apply NLP for a prediction task.

24 December 2022 in TechRxiv
Verified
Preprint
Open access
PDF file
CMChérubin Mugisha
Chérubin Mugisha
Incheon PaikIncheon Paik
Incheon Paik
Specific-domain language models (LMs) are ubiquitous in natural language processing. Effective biomedical text mining requires in-domain models of biomedical and clinical documents, which involve particular challenges. Most existing biomedical LMs rely on self-attention mechanisms, which do not scale to clinical-document length because of space complexity and quadratic execution times. In addition, in-domain LMs ignore the importance of a dedicated tokenizer and of learning optimal hyperparameter values for downstream tasks. In this paper, we introduce a model that addresses these three main issues. Inspired by the BigBird architecture model, we extend the embedding size through an attention mechanism that scales linearly with the sequence length to enable the processing of longer sequences, by a factor of up to eight. Our model was trained on biomedical and raw clinical corpora, and the same data were used to build a domain-specific sentence-piece tokenizer. In our experiments comparing our tokenizer with those in existing models, it was found to have the lowest fertility rate. To optimize the model performance, additional data-oriented hyperparameter fine-tuning was performed to enhance the accuracy and reproducibility of the model. Using the BLURB benchmarks, we evaluated our model with respect to named-entity recognition, relation extraction, sentence similarity, and question-answering tasks. Our model demonstrated competitive results, outperforming other strong self-attention baseline models for several datasets.

1 January 2026 in IEEE Access
Verified
Journal Article
Open access
HDHong N. Dao
Hong N. Dao
TNTuyen Nguyen
Tuyen Nguyen
CMChérubin Mugisha
Chérubin Mugisha
Incheon PaikIncheon Paik
Incheon Paik