Boston University
Compressed encoding and classification of pediatric lung ultrasound data for consolidation detection
Abstract
dc:description.abstractIn this study, we present a lightweight and interpretable method for detecting pulmonary consolidations in pediatric lung ultrasound (LUS) images by compressing two-dimensional (2D) image data into one-dimensional (1D) Brightness Profiles and classifying them using small Multilayer Perceptrons (MLPs). Each LUS frame was transformed into three types of 1D signals— row-wise, column-wise, and concatenated pixel intensity sums—capturing essential brightness intensity variations linked to pathological findings. These profiles served as input to separate MLPs sharing a common architecture. A leave-one-patient-out cross-validation ensured robust evaluation across the dataset. Our 1D+MLP framework matched the performance of conventional CNN-based approaches (Thompson et al., 2023) while significantly lowering computational demands and model pretraining needs.To move from frame-level to video-level decisions, we tested aggregation strategies including majority voting, averaging, and thresholding—the last of which yielded the best classification results. External validation on an independent dataset demonstrated strong generalization. These findings affirm that 1D compressed representations of LUS data, combined with efficient MLPs, offer a viable alternative to CNN-based methods, particularly in low-resource or real-time clinical environments.
Degree
thesis:*- Name thesis:degree_name
- Master of Science
- Level thesis:degree_level
- masters
- Discipline thesis:degree_discipline
- Computer Science
- Grantor
- Boston University
- Year dc:date.issued
- 2025
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Jain, Srishti
- Advisor dc:contributor.advisor
-
- Betke, Margrit
- ORCID dc:identifier.orcid
Subjects
dc:subject × 1Rights
dc:rights- Statement dc:rights
-
- Attribution-NonCommercial-ShareAlike 4.0 International
- Licence dc:rights.uri
- Language dc:language.iso
- en_US
Identifiers
dc:identifier.*- Handle dc:identifier.uri
- https://hdl.handle.net/2144/53066