Clemence Mottez

All Publications

• From Detection to Mitigation: Addressing Bias in Deep Learning Models for Chest X-Ray Diagnosis. Pacific Symposium on Biocomputing. Pacific Symposium on Biocomputing Mottez, C., Fay, L., Varma, M., Ostmeier, S., Langlotz, C. 2026; 31: 538-550

  Abstract

  Deep learning models have shown promise in improving diagnostic accuracy from chest X-rays, but they also risk perpetuating healthcare disparities when performance varies across demographic groups. In this work, we present a comprehensive bias detection and mitigation framework targeting sex, age, and race-based disparities when performing diagnostic tasks with chest X-rays. We extend a recent CNN-XGBoost pipeline to support multi-label classification and evaluate its performance across four medical conditions. We show that replacing the final layer of CNN with an eXtreme Gradient Boosting classifier improves the fairness of the subgroup while maintaining or improving the overall predictive performance. To validate its generalizability, we apply the method to different backbones, namely DenseNet-121 and ResNet-50, and achieve similarly strong performance and fairness outcomes, confirming its model-agnostic design. We further compare this lightweight adapter training method with traditional full-model training bias mitigation techniques, including adversarial training, reweighting, data augmentation, and active learning, and find that our approach offers competitive or superior bias reduction at a fraction of the computational cost. Finally, we show that combining eXtreme Gradient Boosting retraining with active learning yields the largest reduction in bias across all demographic subgroups, both in and out of distribution on the CheXpert and MIMIC datasets, establishing a practical and effective path toward equitable deep learning deployment in clinical radiology.

  View details for DOI 10.1142/9789819824755_0039

  View details for PubMedID 41758167

• A Simplified and Data-Driven Lung Ultrasound Approach for Predicting Surfactant Need in Preterm Infants: A Pilot Study Using Machine Learning and Rule-Based Models. Journal of ultrasound in medicine : official journal of the American Institute of Ultrasound in Medicine Meau-Petit, V., Mottez, C., Bhojnagarwala, B., Montasser, M., Singh, Y., Loganathan, P. K. 2025

  Abstract

  Six-region lung ultrasound (LUS) scores show good predictive value for predicting surfactant need in preterm infants but rely on a fixed threshold, which may lead to misclassification near the cut-off and lack data-driven justification for selecting these 6 regions. This study explored whether evaluating individual regions-and combinations-could improve predictive accuracy and utility.Data from preterm infants born at ≤34 weeks and enrolled in the Serial Lung Ultrasound for Surfactant Replacement Therapy (SLURP) cohort study were analyzed to develop predictive models for surfactant administration based on regional LUS scores. Univariate, bivariate, and machine learning analyses were conducted to identify the most informative lung regions. Rule-based, decision tree, and logistic regression models were then developed, compared to the 6-region model, and validated on an external dataset.The training set consisted of 77 patients from the SLURP cohort study. The rule-based, decision tree, and logistic regression models showed the best performance, primarily using 2 lung regions-left lateral and left upper posterior. A refined model that included the right upper anterior (RUA) region further improved performance. On the external test set (n = 42), the rule-based model with RUA achieved the highest accuracy (0.93) and the lowest false negative rate (0.11), outperforming the 6-region model. Adding more regions did not enhance accuracy.A simplified, rule-based model that accounts for the differential predictive value of individual lung regions may enhance the accuracy of LUS-based prediction of surfactant need in preterm infants. It is also more accessible, effective, and time-efficient for clinicians.

  View details for DOI 10.1002/jum.70092

  View details for PubMedID 41104708