AUTOEUS: Smart Detection of Gastrointestinal Abnormalities Using Lightweight Deep Learning
Fuente:
Zenodo
Enregistré dans:
| Auteur principal: | |
|---|---|
| Format: | Recurso digital |
| Publié: |
Zenodo
2025
|
| Sujets: | |
| Accès en ligne: | |
| Tags: |
Ajouter un tag
Pas de tags, Soyez le premier à ajouter un tag!
|
| _version_ | 1866901539779510272 |
|---|---|
| author | Done. Rama Shanthi*, I. Kullayamma |
| author_facet | Done. Rama Shanthi*, I. Kullayamma |
| contents | <p><span>It is important to accurately identify gastrointestinal (GI) diseases such as adenocarcinomas using Endoscopic Ultrasound (EUS) imaging to attain a timely diagnosis and ultimately better clinical outcomes. However, EUS images present challenges in terms of significant inherent noise, low contrast in color, and complex textural elements, thus making automated analysis a significant challenge toward reliability. Here we present AUTOEUS, a lightweight deep learning framework that improves upon prior models and is able to perform both anatomical region classification and detect adenocarcinomas from EUS images. The proposed pipeline includes a two-stage preprocessing procedure that implements median filtering for noise reduction and Y-channel histogram equalization for contrast improvement, leading to improved clarity in EUS images. In addition, a teacher/student knowledge distillation architecture has been deployed whereby a teacher network uses ResNet-50 to progressively improve the model as a compact convolutional student is guided, thus reducing the computational cost of the model while maintaining predictive accuracy. Experimental evaluations completed within MATLAB using augmented image datastores and five-fold binary classification metrics led to sound diagnostic performance (90.70% - cecum; 95.81% - ileum; 80.00% - pylorus; 90.23% - rectum; 90.23% - stomach). In general F1-scores also reached up to 97.06%. These performance metrics and visualization results confirm the model’s ability to diagnose true positive disease cases with confidence across a number of adenocarcinoma types</span><span>.</span></p> |
| format | Recurso digital |
| id | zenodo_https___doi_org_10_5281_zenodo_17876958 |
| institution | Zenodo |
| language | |
| publishDate | 2025 |
| publisher | Zenodo |
| record_format | zenodo |
| spellingShingle | AUTOEUS: Smart Detection of Gastrointestinal Abnormalities Using Lightweight Deep Learning Done. Rama Shanthi*, I. Kullayamma Endoscopic ultrasound (EUS), Deep learning, Image classification, Knowledge distillation, Convolutional neural network etc <p><span>It is important to accurately identify gastrointestinal (GI) diseases such as adenocarcinomas using Endoscopic Ultrasound (EUS) imaging to attain a timely diagnosis and ultimately better clinical outcomes. However, EUS images present challenges in terms of significant inherent noise, low contrast in color, and complex textural elements, thus making automated analysis a significant challenge toward reliability. Here we present AUTOEUS, a lightweight deep learning framework that improves upon prior models and is able to perform both anatomical region classification and detect adenocarcinomas from EUS images. The proposed pipeline includes a two-stage preprocessing procedure that implements median filtering for noise reduction and Y-channel histogram equalization for contrast improvement, leading to improved clarity in EUS images. In addition, a teacher/student knowledge distillation architecture has been deployed whereby a teacher network uses ResNet-50 to progressively improve the model as a compact convolutional student is guided, thus reducing the computational cost of the model while maintaining predictive accuracy. Experimental evaluations completed within MATLAB using augmented image datastores and five-fold binary classification metrics led to sound diagnostic performance (90.70% - cecum; 95.81% - ileum; 80.00% - pylorus; 90.23% - rectum; 90.23% - stomach). In general F1-scores also reached up to 97.06%. These performance metrics and visualization results confirm the model’s ability to diagnose true positive disease cases with confidence across a number of adenocarcinoma types</span><span>.</span></p> |
| title | AUTOEUS: Smart Detection of Gastrointestinal Abnormalities Using Lightweight Deep Learning |
| topic | Endoscopic ultrasound (EUS), Deep learning, Image classification, Knowledge distillation, Convolutional neural network etc |
| url | https://doi.org/10.5281/zenodo.17876958 |