A Custom Deep Learning Architecture with Image Augmentation for Intelligent Gastrointestinal Tract Tissue Classification
The origin of sampled tissue and characteristics of healthy organs is important for potential abnormality detection in veterinary medicine. Most often such information is given during the sampling process, but in some cases there is a possibility of mislabeling, especially in the education sector wh...
Saved in:
| Main Authors: | , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Sciendo
2024-09-01
|
| Series: | International Journal of Applied Mathematics and Computer Science |
| Subjects: | |
| Online Access: | https://doi.org/10.61822/amcs-2024-0040 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1850033784709185536 |
|---|---|
| author | Wieczorek Michał Wojtas Natalia Wituła Roman Krawczyk Aleksandra Rycerz Karol |
| author_facet | Wieczorek Michał Wojtas Natalia Wituła Roman Krawczyk Aleksandra Rycerz Karol |
| author_sort | Wieczorek Michał |
| collection | DOAJ |
| description | The origin of sampled tissue and characteristics of healthy organs is important for potential abnormality detection in veterinary medicine. Most often such information is given during the sampling process, but in some cases there is a possibility of mislabeling, especially in the education sector where some microscopic preparations might be made by students without proper knowledge or the inspected tissues are not of fully known origin. Occasionally, it is possible to determine the affiliation by searching characteristics of the organ in a sample; however, this is not always possible even for a skilled professional as some tissues vary too little between their counterparts in different species or even in different organs of the same species. Because of this, an automatic system able to perform such classification in a fraction of a second and with high accuracy can be helpful in such cases, especially considering the low cost of adding that solution to the current workflow. This paper presents a new dataset for healthy organ classification based on light microscope imagery containing 25 abstract classes of different organs of a few species. During the sampling process, 3680 images of healthy tissues were collected. Additionally, a custom deep learning architecture was created that is able to classify those samples between organs and species with a validation accuracy reaching 98.34%. Such performance is in some cases higher than that of a human specialist, especially when some examples have very small visual differences between one another or the classification is made on previously non-determining regions of the organ. Additionally, the collection of such a dataset provides a great opportunity for further work containing abnormality detection as it already provides information on the healthy organ description, which can be used for a deep learning model searching for illnesses or mutations. What is more, such a dataset and the corresponding artificial neural network constitute one of the first solutions of this kind in veterinary medicine, as most state-of-the-art papers focus on human medicine. |
| format | Article |
| id | doaj-art-39d3ce0f9a2f46b499ab0d6757afb906 |
| institution | DOAJ |
| issn | 2083-8492 |
| language | English |
| publishDate | 2024-09-01 |
| publisher | Sciendo |
| record_format | Article |
| series | International Journal of Applied Mathematics and Computer Science |
| spelling | doaj-art-39d3ce0f9a2f46b499ab0d6757afb9062025-08-20T02:58:04ZengSciendoInternational Journal of Applied Mathematics and Computer Science2083-84922024-09-0134459761610.61822/amcs-2024-0040A Custom Deep Learning Architecture with Image Augmentation for Intelligent Gastrointestinal Tract Tissue ClassificationWieczorek Michał0Wojtas Natalia1Wituła Roman2Krawczyk Aleksandra3Rycerz Karol4Faculty of Applied Mathematics Silesian University of Technology Kaszubska 23, 44-100 Gliwice, PolandPhD student University of Life Sciences in Lublin Akademicka 12, 20-950 Lublin, PolandFaculty of Applied Mathematics Silesian University of Technology Kaszubska 23, 44-100 Gliwice, PolandDepartment of Animal Anatomy and Histology University of Life Sciences in Lublin Akademicka 12, 20-950 Lublin, PolandDepartment of Animal Anatomy and Histology University of Life Sciences in Lublin Akademicka 12, 20-950 Lublin, PolandThe origin of sampled tissue and characteristics of healthy organs is important for potential abnormality detection in veterinary medicine. Most often such information is given during the sampling process, but in some cases there is a possibility of mislabeling, especially in the education sector where some microscopic preparations might be made by students without proper knowledge or the inspected tissues are not of fully known origin. Occasionally, it is possible to determine the affiliation by searching characteristics of the organ in a sample; however, this is not always possible even for a skilled professional as some tissues vary too little between their counterparts in different species or even in different organs of the same species. Because of this, an automatic system able to perform such classification in a fraction of a second and with high accuracy can be helpful in such cases, especially considering the low cost of adding that solution to the current workflow. This paper presents a new dataset for healthy organ classification based on light microscope imagery containing 25 abstract classes of different organs of a few species. During the sampling process, 3680 images of healthy tissues were collected. Additionally, a custom deep learning architecture was created that is able to classify those samples between organs and species with a validation accuracy reaching 98.34%. Such performance is in some cases higher than that of a human specialist, especially when some examples have very small visual differences between one another or the classification is made on previously non-determining regions of the organ. Additionally, the collection of such a dataset provides a great opportunity for further work containing abnormality detection as it already provides information on the healthy organ description, which can be used for a deep learning model searching for illnesses or mutations. What is more, such a dataset and the corresponding artificial neural network constitute one of the first solutions of this kind in veterinary medicine, as most state-of-the-art papers focus on human medicine.https://doi.org/10.61822/amcs-2024-0040deep learningartificial intelligenceimage augmentationtissue classificationveterinary medicine. |
| spellingShingle | Wieczorek Michał Wojtas Natalia Wituła Roman Krawczyk Aleksandra Rycerz Karol A Custom Deep Learning Architecture with Image Augmentation for Intelligent Gastrointestinal Tract Tissue Classification International Journal of Applied Mathematics and Computer Science deep learning artificial intelligence image augmentation tissue classification veterinary medicine. |
| title | A Custom Deep Learning Architecture with Image Augmentation for Intelligent Gastrointestinal Tract Tissue Classification |
| title_full | A Custom Deep Learning Architecture with Image Augmentation for Intelligent Gastrointestinal Tract Tissue Classification |
| title_fullStr | A Custom Deep Learning Architecture with Image Augmentation for Intelligent Gastrointestinal Tract Tissue Classification |
| title_full_unstemmed | A Custom Deep Learning Architecture with Image Augmentation for Intelligent Gastrointestinal Tract Tissue Classification |
| title_short | A Custom Deep Learning Architecture with Image Augmentation for Intelligent Gastrointestinal Tract Tissue Classification |
| title_sort | custom deep learning architecture with image augmentation for intelligent gastrointestinal tract tissue classification |
| topic | deep learning artificial intelligence image augmentation tissue classification veterinary medicine. |
| url | https://doi.org/10.61822/amcs-2024-0040 |
| work_keys_str_mv | AT wieczorekmichał acustomdeeplearningarchitecturewithimageaugmentationforintelligentgastrointestinaltracttissueclassification AT wojtasnatalia acustomdeeplearningarchitecturewithimageaugmentationforintelligentgastrointestinaltracttissueclassification AT witułaroman acustomdeeplearningarchitecturewithimageaugmentationforintelligentgastrointestinaltracttissueclassification AT krawczykaleksandra acustomdeeplearningarchitecturewithimageaugmentationforintelligentgastrointestinaltracttissueclassification AT rycerzkarol acustomdeeplearningarchitecturewithimageaugmentationforintelligentgastrointestinaltracttissueclassification AT wieczorekmichał customdeeplearningarchitecturewithimageaugmentationforintelligentgastrointestinaltracttissueclassification AT wojtasnatalia customdeeplearningarchitecturewithimageaugmentationforintelligentgastrointestinaltracttissueclassification AT witułaroman customdeeplearningarchitecturewithimageaugmentationforintelligentgastrointestinaltracttissueclassification AT krawczykaleksandra customdeeplearningarchitecturewithimageaugmentationforintelligentgastrointestinaltracttissueclassification AT rycerzkarol customdeeplearningarchitecturewithimageaugmentationforintelligentgastrointestinaltracttissueclassification |