Search research articles

ABOUT JoVE

Overview Leadership Blog JoVE Help Center

AUTHORS

Publishing Process Editorial Board Scope & Policies Peer Review FAQ Submit

LIBRARIANS

Testimonials Subscriptions Access Resources Library Advisory Board FAQ

RESEARCH

JoVE Journal Methods Collections JoVE Encyclopedia of Experiments Archive

EDUCATION

JoVE Core JoVE Business JoVE Science Education JoVE Lab Manual Faculty Resource Center Faculty Site

Terms & Conditions of Use

Search research articles

Related Experiment Video

Updated: Jan 18, 2026

Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography

Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography

Published on: November 30, 2022

Dimensionality Reduction and Nearest Neighbors for Improving Out-of-Distribution Detection in Medical Image

McKell Woodland¹, Nihil Patel¹, Austin Castelo¹

¹The University of Texas MD Anderson Cancer Center, Houston, TX, USA, Rice University, Houston, TX, USA.

The Journal of Machine Learning for Biomedical Imaging

|June 2, 2025

Summary

Related Concept Videos

You might also read

Related Articles

Articles linked to this work by shared authors, journal, and citation graph.

Sort by

Same author

RADIANT: A fully configurable radiotherapy dose prediction framework.

Biomedical physics & engineering express·2026

Same author

Clinicogenomic and Histopathologic Analyses of Supermassive Intrahepatic Cholangiocarcinoma and the Role of Ablative Radiotherapy.

Clinical cancer research : an official journal of the American Association for Cancer Research·2026

Same author

Quantifying Functional Liver Volume Loss after CT-guided Percutaneous Thermal Ablation: A COVER-ALL Trial Post Hoc Analysis.

Radiology. Imaging cancer·2026

Same author

Reply to: Methodologic Considerations for Subsequent Colorectal Cancer in Survivors of Childhood Cancer.

Journal of clinical oncology : official journal of the American Society of Clinical Oncology·2026

Same author

Molecular Profiling of Cholangiocarcinoma Predicts Outcomes Post-Liver Transplantation.

JCO precision oncology·2026

Same author

Quality versus quantity of training datasets for artificial intelligence-based whole liver segmentation.

medRxiv : the preprint server for health sciences·2026

Same journal

Finding Reproducible and Prognostic Radiomic Features in Variable Slice Thickness Contrast Enhanced CT of Colorectal Liver Metastases.

The journal of machine learning for biomedical imaging·2025

Same journal

A Neural Conditional Random Field Model Using Deep Features and Learnable Functions for End-to-End MRI Prostate Zonal Segmentation.

The journal of machine learning for biomedical imaging·2025

Same journal

Leveraging SO(3)-steerable convolutions for pose-robust semantic segmentation in 3D medical data.

The journal of machine learning for biomedical imaging·2024

Same journal

Shape-aware Segmentation of the Placenta in BOLD Fetal MRI Time Series.

The journal of machine learning for biomedical imaging·2024

Same journal

Compound Figure Separation of Biomedical Images: Mining Large Datasets for Self-supervised Learning.

The journal of machine learning for biomedical imaging·2023

Same journal

QU-BraTS: MICCAI BraTS 2020 Challenge on Quantifying Uncertainty in Brain Tumor Segmentation - Analysis of Ranking Scores and Benchmarking Results.

The journal of machine learning for biomedical imaging·2023

See all related articles

This summary is machine-generated.

Detecting out-of-distribution images is crucial for deep learning models in clinical settings. This study shows dimension reduction techniques combined with Mahalanobis distance or k-nearest neighbors effectively identify model failures.

Area of Science:

Medical Imaging
Artificial Intelligence
Machine Learning

Background:

Deep learning segmentation models can fail on data outside their training distribution.
Automation bias may arise from reliable performance on most cases, masking occasional failures.
Detecting out-of-distribution (OOD) images during inference is critical to alert clinicians to potential model errors.

Purpose of the Study:

To evaluate dimension reduction techniques for detecting OOD images in deep learning segmentation models.
To assess the performance of Mahalanobis distance (MD) and k-nearest neighbors (KNN) for OOD detection.
To improve the reliability of deep learning models in clinical practice by identifying failure cases.

Main Methods:

Applied Mahalanobis distance (MD) to bottleneck features of Swin UNETR and nnU-net models segmenting liver MRIs and CTs.

Keywords:

Mahalanobis distance Nearest Neighbors Out-of-distribution detection Principal component analysis Uniform manifold approximation and projection

Related Experiment Videos

Last Updated: Jan 18, 2026

Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography

Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography

Published on: November 30, 2022

Reduced bottleneck feature dimensions using Principal Component Analysis (PCA) and Uniform Manifold Approximation and Projection (UMAP).

Explored k-th nearest neighbors distance (KNN) as a non-parametric alternative to MD on raw and pooled features.

Main Results:

Dimension reduction techniques (PCA, UMAP) enabled high-performance detection of failed segmentations with minimal computational cost.
K-nearest neighbors (KNN) significantly outperformed MD in scalability and detection accuracy when applied to bottleneck features.
Both MD and KNN, when combined with dimension reduction, successfully identified out-of-distribution images where models failed.

Conclusions:

Dimension reduction is effective for enhancing OOD detection in clinical deep learning segmentation.
KNN offers a scalable and high-performing alternative to MD for OOD detection in medical imaging AI.
Implementing these methods can mitigate automation bias and improve the safety of AI in radiology.