Rita Noumeir

Aamer Abdul Rahman, Pranav Agarwal, Rita Noumeir et al. · mila

Offline reinforcement learning has shown promise for solving tasks in safety-critical settings, such as clinical decision support. Its application, however, has been limited by the lack of interpretability and interactivity for clinicians. To address these challenges, we propose the medical decision transformer (MeDT), a novel and versatile framework based on the goal-conditioned reinforcement learning paradigm for sepsis treatment recommendation. MeDT uses the decision transformer architecture to learn a policy for drug dosage recommendation. During offline training, MeDT utilizes collected treatment trajectories to predict administered treatments for each time step, incorporating known treatment outcomes, target acuity scores, past treatment decisions, and current and past medical states. This analysis enables MeDT to capture complex dependencies among a patient's medical history, treatment decisions, outcomes, and short-term effects on stability. Our proposed conditioning uses acuity scores to address sparse reward issues and to facilitate clinician-model interactions, enhancing decision-making. Following training, MeDT can generate tailored treatment recommendations by conditioning on the desired positive outcome (survival) and user-specified short-term stability improvements. We carry out rigorous experiments on data from the MIMIC-III dataset and use off-policy evaluation to demonstrate that MeDT recommends interventions that outperform or are competitive with existing offline reinforcement learning methods while enabling a more interpretable, personalized and clinician-directed approach.

Thanh-Dung Le, Rita Noumeir, Jerome Rambaud et al.

When dealing with clinical text classification on a small dataset recent studies have confirmed that a well-tuned multilayer perceptron outperforms other generative classifiers, including deep learning ones. To increase the performance of the neural network classifier, feature selection for the learning representation can effectively be used. However, most feature selection methods only estimate the degree of linear dependency between variables and select the best features based on univariate statistical tests. Furthermore, the sparsity of the feature space involved in the learning representation is ignored. Goal: Our aim is therefore to access an alternative approach to tackle the sparsity by compressing the clinical representation feature space, where limited French clinical notes can also be dealt with effectively. Methods: This study proposed an autoencoder learning algorithm to take advantage of sparsity reduction in clinical note representation. The motivation was to determine how to compress sparse, high-dimensional data by reducing the dimension of the clinical note representation feature space. The classification performance of the classifiers was then evaluated in the trained and compressed feature space. Results: The proposed approach provided overall performance gains of up to 3% for each evaluation. Finally, the classifier achieved a 92% accuracy, 91% recall, 91% precision, and 91% f1-score in detecting the patient's condition. Furthermore, the compression working mechanism and the autoencoder prediction process were demonstrated by applying the theoretic information bottleneck framework.

Thanh-Dung Le, Ti Ti Nguyen, Vu Nguyen Ha et al.

Fine-tuning Large Language Models (LLMs) for clinical Natural Language Processing (NLP) poses significant challenges due to domain gap, limited data, and stringent hardware constraints. In this study, we evaluate four adapter techniques-Adapter, Lightweight, TinyAttention, and Gated Residual Network (GRN) - equivalent to Low-Rank Adaptation (LoRA), for clinical note classification under real-world, resource-constrained conditions. All experiments were conducted on a single NVIDIA Quadro P620 GPU (2 GB VRAM, 512 CUDA cores, 1.386 TFLOPS FP32), limiting batch sizes to <8 sequences and maximum sequence length to 256 tokens. Our clinical corpus comprises only 580 000 tokens, several orders of magnitude smaller than standard LLM pre-training datasets. We fine-tuned three biomedical pre-trained LLMs (CamemBERT-bio, AliBERT, DrBERT) and two lightweight Transformer models trained from scratch. Results show that 1) adapter structures provide no consistent gains when fine-tuning biomedical LLMs under these constraints, and 2) simpler Transformers, with minimal parameter counts and training times under six hours, outperform adapter-augmented LLMs, which required over 1000 GPU-hours. Among adapters, GRN achieved the best metrics (accuracy, precision, recall, F1 = 0.88). These findings demonstrate that, in low-resource clinical settings with limited data and compute, lightweight Transformers trained from scratch offer a more practical and efficient solution than large LLMs, while GRN remains a viable adapter choice when minimal adaptation is needed.

Thanh-Dung Le, Philippe Jouvet, Rita Noumeir

Transformer-based models have shown outstanding results in natural language processing but face challenges in applications like classifying small-scale clinical texts, especially with constrained computational resources. This study presents a customized Mixture of Expert (MoE) Transformer models for classifying small-scale French clinical texts at CHU Sainte-Justine Hospital. The MoE-Transformer addresses the dual challenges of effective training with limited data and low-resource computation suitable for in-house hospital use. Despite the success of biomedical pre-trained models such as CamemBERT-bio, DrBERT, and AliBERT, their high computational demands make them impractical for many clinical settings. Our MoE-Transformer model not only outperforms DistillBERT, CamemBERT, FlauBERT, and Transformer models on the same dataset but also achieves impressive results: an accuracy of 87\%, precision of 87\%, recall of 85\%, and F1-score of 86\%. While the MoE-Transformer does not surpass the performance of biomedical pre-trained BERT models, it can be trained at least 190 times faster, offering a viable alternative for settings with limited data and computational resources. Although the MoE-Transformer addresses challenges of generalization gaps and sharp minima, demonstrating some limitations for efficient and accurate clinical text classification, this model still represents a significant advancement in the field. It is particularly valuable for classifying small French clinical narratives within the privacy and constraints of hospital-based computational resources.

Clara Macabiau, Thanh-Dung Le, Kevin Albert et al.

This study aimed to investigate the application of label propagation techniques to propagate labels among photoplethysmogram (PPG) signals, particularly in imbalanced class scenarios and limited data availability scenarios, where clean PPG samples are significantly outnumbered by artifact-contaminated samples. We investigated a dataset comprising PPG recordings from 1571 patients, wherein approximately 82% of the samples were identified as clean, while the remaining 18% were contaminated by artifacts. Our research compares the performance of supervised classifiers, such as conventional classifiers and neural networks (Multi-Layer Perceptron (MLP), Transformers, Fully Convolutional Network (FCN)), with the semi-supervised Label Propagation (LP) algorithm for artifact classification in PPG signals. The results indicate that the LP algorithm achieves a precision of 91%, a recall of 90%, and an F1 score of 90% for the "artifacts" class, showcasing its effectiveness in annotating a medical dataset, even in cases where clean samples are rare. Although the K-Nearest Neighbors (KNN) supervised model demonstrated good results with a precision of 89%, a recall of 95%, and an F1 score of 92%, the semi-supervised algorithm excels in artifact detection. In the case of imbalanced and limited pediatric intensive care environment data, the semi-supervised LP algorithm is promising for artifact detection in PPG signals. The results of this study are important for improving the accuracy of PPG-based health monitoring, particularly in situations in which motion artifacts pose challenges to data interpretation

Mohamed Khalil Ben Salah, Philippe Jouvet, Rita Noumeir

Continuous monitoring of vital signs in Pediatric Intensive Care Units (PICUs) is essential for early detection of clinical deterioration and effective clinical decision-making. However, contact-based sensors such as pulse oximeters may cause skin irritation, increase infection risk, and lead to patient discomfort. Remote photoplethysmography (rPPG) offers a contactless alternative to monitor heart rate using facial video, but remains underutilized in PICUs due to motion artifacts, occlusions, variable lighting, and domain shifts between laboratory and clinical data. We introduce a self-supervised pretraining framework for rPPG estimation in the PICU setting, based on a progressive curriculum strategy. The approach leverages the VisionMamba architecture and integrates an adaptive masking mechanism, where a lightweight Mamba-based controller assigns spatiotemporal importance scores to guide probabilistic patch sampling. This strategy dynamically increases reconstruction difficulty while preserving physiological relevance. To address the lack of labeled clinical data, we adopt a teacher-student distillation setup. A supervised expert model, trained on public datasets, provides latent physiological guidance to the student. The curriculum progresses through three stages: clean public videos, synthetic occlusion scenarios, and unlabeled videos from 500 pediatric patients. Our framework achieves a 42% reduction in mean absolute error relative to standard masked autoencoders and outperforms PhysFormer by 31%, reaching a final MAE of 3.2 bpm. Without explicit region-of-interest extraction, the model consistently attends to pulse-rich areas and demonstrates robustness under clinical occlusions and noise.

Thanh-Dung Le, Clara Macabiau, Kévin Albert et al.

Recent research at CHU Sainte Justine's Pediatric Critical Care Unit (PICU) has revealed that traditional machine learning methods, such as semi-supervised label propagation and K-nearest neighbors, outperform Transformer-based models in artifact detection from PPG signals, mainly when data is limited. This study addresses the underutilization of abundant unlabeled data by employing self-supervised learning (SSL) to extract latent features from these data, followed by fine-tuning on labeled data. Our experiments demonstrate that SSL significantly enhances the Transformer model's ability to learn representations, improving its robustness in artifact classification tasks. Among various SSL techniques, including masking, contrastive learning, and DINO (self-distillation with no labels)-contrastive learning exhibited the most stable and superior performance in small PPG datasets. Further, we delve into optimizing contrastive loss functions, which are crucial for contrastive SSL. Inspired by InfoNCE, we introduce a novel contrastive loss function that facilitates smoother training and better convergence, thereby enhancing performance in artifact classification. In summary, this study establishes the efficacy of SSL in leveraging unlabeled data, particularly in enhancing the capabilities of the Transformer model. This approach holds promise for broader applications in PICU environments, where annotated data is often limited.

Thanh-Dung Le, Rita Noumeir, Jerome Rambaud et al.

The purpose of the study presented herein is to develop a machine learning algorithm based on natural language processing that automatically detects whether a patient has a cardiac failure or a healthy condition by using physician notes in Research Data Warehouse at CHU Sainte Justine Hospital. First, a word representation learning technique was employed by using bag-of-word (BoW), term frequency inverse document frequency (TFIDF), and neural word embeddings (word2vec). Each representation technique aims to retain the words semantic and syntactic analysis in critical care data. It helps to enrich the mutual information for the word representation and leads to an advantage for further appropriate analysis steps. Second, a machine learning classifier was used to detect the patients condition for either cardiac failure or stable patient through the created word representation vector space from the previous step. This machine learning approach is based on a supervised binary classification algorithm, including logistic regression (LR), Gaussian Naive-Bayes (GaussianNB), and multilayer perceptron neural network (MLPNN). Technically, it mainly optimizes the empirical loss during training the classifiers. As a result, an automatic learning algorithm would be accomplished to draw a high classification performance, including accuracy (acc), precision (pre), recall (rec), and F1 score (f1). The results show that the combination of TFIDF and MLPNN always outperformed other combinations with all overall performance. In the case without any feature selection, the proposed framework yielded an overall classification performance with acc, pre, rec, and f1 of 84% and 82%, 85%, and 83%, respectively. Significantly, if the feature selection was well applied, the overall performance would finally improve up to 4% for each evaluation.

Thanh-Dung Le, Rita Noumeir, Jerome Rambaud et al.

The rapid progress in clinical data management systems and artificial intelligence approaches enable the era of personalized medicine. Intensive care units (ICUs) are the ideal clinical research environment for such development because they collect many clinical data and are highly computerized environments. We designed a retrospective clinical study on a prospective ICU database using clinical natural language to help in the early diagnosis of heart failure in critically ill children. The methodology consisted of empirical experiments of a learning algorithm to learn the hidden interpretation and presentation of the French clinical note data. This study included 1386 patients' clinical notes with 5444 single lines of notes. There were 1941 positive cases (36 % of total) and 3503 negative cases classified by two independent physicians using a standardized approach. The multilayer perceptron neural network outperforms other discriminative and generative classifiers. Consequently, the proposed framework yields an overall classification performance with 89 % accuracy, 88 % recall, and 89 % precision. This study successfully applied learning representation and machine learning algorithms to detect heart failure from clinical natural language in a single French institution. Further work is needed to use the same methodology in other institutions and other languages.

Alban Main de Boissiere, Rita Noumeir

Action recognition, early prediction, and online action detection are complementary disciplines that are often studied independently. Most online action detection networks use a pre-trained feature extractor, which might not be optimal for its new task. We address the task-specific feature extraction with a teacher-student framework between the aforementioned disciplines, and a novel training strategy. Our network, Online Knowledge Distillation Action Detection network (OKDAD), embeds online early prediction and online temporal segment proposal subnetworks in parallel. Low interclass and high intraclass similarity are encouraged during teacher training. Knowledge distillation to the OKDAD network is ensured via layer reuse and cosine similarity between teacher-student feature vectors. Layer reuse and similarity learning significantly improve our baseline which uses a generic feature extractor. We evaluate our framework on infrared videos from two popular datasets, NTU RGB+D (action recognition, early prediction) and PKU MMD (action detection). Unlike previous attempts on those datasets, our student networks perform without any knowledge of the future. Even with this added difficulty, we achieve state-of-the-art results on both datasets. Moreover, our networks use infrared from RGB-D cameras, which we are the first to use for online action detection, to our knowledge.

Alban Main de Boissiere, Rita Noumeir

A challenge of skeleton-based action recognition is the difficulty to classify actions with similar motions and object-related actions. Visual clues from other streams help in that regard. RGB data are sensible to illumination conditions, thus unusable in the dark. To alleviate this issue and still benefit from a visual stream, we propose a modular network (FUSION) combining skeleton and infrared data. A 2D convolutional neural network (CNN) is used as a pose module to extract features from skeleton data. A 3D CNN is used as an infrared module to extract visual cues from videos. Both feature vectors are then concatenated and exploited conjointly using a multilayer perceptron (MLP). Skeleton data also condition the infrared videos, providing a crop around the performing subjects and thus virtually focusing the attention of the infrared module. Ablation studies show that using pre-trained networks on other large scale datasets as our modules and data augmentation yield considerable improvements on the action classification accuracy. The strong contribution of our cropping strategy is also demonstrated. We evaluate our method on the NTU RGB+D dataset, the largest dataset for human action recognition from depth cameras, and report state-of-the-art performances.