Abstract:The integration of deep learning tools in gastrointestinal vision holds the potential for significant advancements in diagnosis, treatment, and overall patient care. A major challenge, however, is these tools' tendency to make overconfident predictions, even when encountering unseen or newly emerging disease patterns, undermining their reliability. We address this critical issue of reliability by framing it as an out-of-distribution (OOD) detection problem, where previously unseen and emerging diseases are identified as OOD examples. However, gastrointestinal images pose a unique challenge due to the overlapping feature representations between in- Distribution (ID) and OOD examples. Existing approaches often overlook this characteristic, as they are primarily developed for natural image datasets, where feature distinctions are more apparent. Despite the overlap, we hypothesize that the features of an in-distribution example will cluster closer to the centroids of their ground truth class, resulting in a shorter distance to the nearest centroid. In contrast, OOD examples maintain an equal distance from all class centroids. Based on this observation, we propose a novel nearest-centroid distance deficit (NCCD) score in the feature space for gastrointestinal OOD detection. Evaluations across multiple deep learning architectures and two publicly available benchmarks, Kvasir2 and Gastrovision, demonstrate the effectiveness of our approach compared to several state-of-the-art methods. The code and implementation details are publicly available at: <a class="link-external link-https" href="https://github.com/bhattarailab/NCDD" rel="external noopener nofollow">this https URL</a>

What problem does this paper attempt to address?

### What problems does this paper attempt to solve? This paper aims to solve the problem of over - confident predictions made by deep - learning tools when encountering unseen or newly - emerging disease patterns in gastrointestinal vision. Specifically: 1. **Reliability issue**: Existing deep - learning models perform poorly in handling open - set classification tasks (such as detecting rare and new anomalies), especially in the medical field. These models do not fully consider rare diseases and unseen abnormal situations, resulting in over - confident predictions for unknown cases, thus affecting their reliability and credibility. 2. **Feature overlap issue**: There is an overlap in feature representation in gastrointestinal images, making it more difficult to distinguish between in - distribution (ID) and out - of - distribution (OOD) samples. Most of the existing methods are developed for natural image datasets, where the feature differences are more obvious, while the feature overlap problem in gastrointestinal images is ignored. To solve these problems, the author redefines this challenge as an OOD detection problem and proposes a new method based on Nearest Centroid Distance Deficit (NCDD). Specifically: - **Research objective**: By introducing the NCDD score and using the distance information of the nearest class cluster centers in the feature space to distinguish between ID and OOD samples, the reliability of the model in detecting unseen or newly - emerging disease patterns is improved. - **Innovations**: - Redefining the anomaly detection problem as an OOD problem, allowing the model to recognize pathological features without explicit training. - Proposing a new distance - based OOD detection method NCDD, which combines the information of the nearest class cluster center and non - nearest class cluster centers. - The method is easy to implement and applicable to multiple model architectures, and it is a posterior method. Through evaluations on multiple deep - learning architectures and public benchmark datasets, the effectiveness of this method is verified, and its superior performance in the OOD detection task of gastrointestinal images is demonstrated.

NCDD: Nearest Centroid Distance Deficit for Out-Of-Distribution Detection in Gastrointestinal Vision

TTA-OOD: Test-time Augmentation for Improving Out-of-Distribution Detection in Gastrointestinal Vision

Out-of-Distribution Detection with Deep Nearest Neighbors

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

DNN-GDITD: Out-of-distribution detection via Deep Neural Network based Gaussian Descriptor for Imbalanced Tabular Data

Unveiling the unseen: novel strategies for object detection beyond known distributions

Are We Ready for Out-of-Distribution Detection in Digital Pathology?

OOD-SEG: Out-Of-Distribution detection for image SEGmentation with sparse multi-class positive-only annotations

Low-Dimensional Gradient Helps Out-of-Distribution Detection

Out-of-distribution Detection in Medical Image Analysis: A survey

GAIA: Delving into Gradient-based Attribution Abnormality for Out-of-distribution Detection

Limitations of Out-of-Distribution Detection in 3D Medical Image Segmentation

Redesigning Out-of-Distribution Detection on 3D Medical Images

Angiodysplasia Detection and Localization Using Deep Convolutional Neural Networks

Dimensionality Reduction for Improving Out-of-Distribution Detection in Medical Image Segmentation

EndoOOD: Uncertainty-aware Out-of-distribution Detection in Capsule Endoscopy Diagnosis

Calibrated Out-of-Distribution Detection with a Generic Representation

FOOD: Fast Out-Of-Distribution Detector

NECO: NEural Collapse Based Out-of-distribution detection

Out-of-Distribution Detection for Long-tailed and Fine-grained Skin Lesion Images

Your Data Is Not Perfect: Towards Cross-Domain Out-of-Distribution Detection in Class-Imbalanced Data