A Systematic Comparison of Task Adaptation Techniques for Digital Histopathology
Due to an insufficient amount of image annotation, artificial intelligence in computational histopathology usually relies on fine-tuning pre-trained neural networks. While vanilla fine-tuning has shown to be effective, research on computer vision has recently proposed improved algorithms, promising...
| Main Authors: | , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
MDPI AG
2023-12-01
|
| Series: | Bioengineering |
| Subjects: | |
| Online Access: | https://www.mdpi.com/2306-5354/11/1/19 |
| _version_ | 1827372768564346880 |
|---|---|
| author | Daniel Sauter Georg Lodde Felix Nensa Dirk Schadendorf Elisabeth Livingstone Markus Kukuk |
| author_facet | Daniel Sauter Georg Lodde Felix Nensa Dirk Schadendorf Elisabeth Livingstone Markus Kukuk |
| author_sort | Daniel Sauter |
| collection | DOAJ |
| description | Due to an insufficient amount of image annotation, artificial intelligence in computational histopathology usually relies on fine-tuning pre-trained neural networks. While vanilla fine-tuning has shown to be effective, research on computer vision has recently proposed improved algorithms, promising better accuracy. While initial studies have demonstrated the benefits of these algorithms for medical AI, in particular for radiology, there is no empirical evidence for improved accuracy in histopathology. Therefore, based on the ConvNeXt architecture, our study performs a systematic comparison of nine task adaptation techniques, namely, DELTA, L<sup>2</sup>-SP, MARS-PGM, Bi-Tuning, BSS, MultiTune, SpotTune, Co-Tuning, and vanilla fine-tuning, on five histopathological classification tasks using eight datasets. The results are based on external testing and statistical validation and reveal a multifaceted picture: some techniques are better suited for histopathology than others, but depending on the classification task, a significant relative improvement in accuracy was observed for five advanced task adaptation techniques over the control method, i.e., vanilla fine-tuning (e.g., Co-Tuning: <i>P</i>(≫) = 0.942, <i>d</i> = 2.623). Furthermore, we studied the classification accuracy for three of the nine methods with respect to the training set size (e.g., Co-Tuning: <i>P</i>(≫) = 0.951, <i>γ</i> = 0.748). Overall, our results show that the performance of advanced task adaptation techniques in histopathology is affected by influencing factors such as the specific classification task or the size of the training dataset. |
| first_indexed | 2024-03-08T11:05:51Z |
| format | Article |
| id | doaj.art-f6767e9b2d44400cb385568a87d03613 |
| institution | Directory Open Access Journal |
| issn | 2306-5354 |
| language | English |
| last_indexed | 2024-03-08T11:05:51Z |
| publishDate | 2023-12-01 |
| publisher | MDPI AG |
| record_format | Article |
| series | Bioengineering |
| spelling | doaj.art-f6767e9b2d44400cb385568a87d036132024-01-26T15:06:08ZengMDPI AGBioengineering2306-53542023-12-011111910.3390/bioengineering11010019A Systematic Comparison of Task Adaptation Techniques for Digital HistopathologyDaniel Sauter0Georg Lodde1Felix Nensa2Dirk Schadendorf3Elisabeth Livingstone4Markus Kukuk5Department of Computer Science, Fachhochschule Dortmund, 44227 Dortmund, GermanyDepartment of Dermatology, University Hospital Essen, 45147 Essen, GermanyInstitute for AI in Medicine (IKIM), University Hospital Essen, 45131 Essen, GermanyDepartment of Dermatology, University Hospital Essen, 45147 Essen, GermanyDepartment of Dermatology, University Hospital Essen, 45147 Essen, GermanyDepartment of Computer Science, Fachhochschule Dortmund, 44227 Dortmund, GermanyDue to an insufficient amount of image annotation, artificial intelligence in computational histopathology usually relies on fine-tuning pre-trained neural networks. While vanilla fine-tuning has shown to be effective, research on computer vision has recently proposed improved algorithms, promising better accuracy. While initial studies have demonstrated the benefits of these algorithms for medical AI, in particular for radiology, there is no empirical evidence for improved accuracy in histopathology. Therefore, based on the ConvNeXt architecture, our study performs a systematic comparison of nine task adaptation techniques, namely, DELTA, L<sup>2</sup>-SP, MARS-PGM, Bi-Tuning, BSS, MultiTune, SpotTune, Co-Tuning, and vanilla fine-tuning, on five histopathological classification tasks using eight datasets. The results are based on external testing and statistical validation and reveal a multifaceted picture: some techniques are better suited for histopathology than others, but depending on the classification task, a significant relative improvement in accuracy was observed for five advanced task adaptation techniques over the control method, i.e., vanilla fine-tuning (e.g., Co-Tuning: <i>P</i>(≫) = 0.942, <i>d</i> = 2.623). Furthermore, we studied the classification accuracy for three of the nine methods with respect to the training set size (e.g., Co-Tuning: <i>P</i>(≫) = 0.951, <i>γ</i> = 0.748). Overall, our results show that the performance of advanced task adaptation techniques in histopathology is affected by influencing factors such as the specific classification task or the size of the training dataset.https://www.mdpi.com/2306-5354/11/1/19transfer learningfine-tuningcomputer visionCNNwhole-slide imagingcancer |
| spellingShingle | Daniel Sauter Georg Lodde Felix Nensa Dirk Schadendorf Elisabeth Livingstone Markus Kukuk A Systematic Comparison of Task Adaptation Techniques for Digital Histopathology Bioengineering transfer learning fine-tuning computer vision CNN whole-slide imaging cancer |
| title | A Systematic Comparison of Task Adaptation Techniques for Digital Histopathology |
| title_full | A Systematic Comparison of Task Adaptation Techniques for Digital Histopathology |
| title_fullStr | A Systematic Comparison of Task Adaptation Techniques for Digital Histopathology |
| title_full_unstemmed | A Systematic Comparison of Task Adaptation Techniques for Digital Histopathology |
| title_short | A Systematic Comparison of Task Adaptation Techniques for Digital Histopathology |
| title_sort | systematic comparison of task adaptation techniques for digital histopathology |
| topic | transfer learning fine-tuning computer vision CNN whole-slide imaging cancer |
| url | https://www.mdpi.com/2306-5354/11/1/19 |
| work_keys_str_mv | AT danielsauter asystematiccomparisonoftaskadaptationtechniquesfordigitalhistopathology AT georglodde asystematiccomparisonoftaskadaptationtechniquesfordigitalhistopathology AT felixnensa asystematiccomparisonoftaskadaptationtechniquesfordigitalhistopathology AT dirkschadendorf asystematiccomparisonoftaskadaptationtechniquesfordigitalhistopathology AT elisabethlivingstone asystematiccomparisonoftaskadaptationtechniquesfordigitalhistopathology AT markuskukuk asystematiccomparisonoftaskadaptationtechniquesfordigitalhistopathology AT danielsauter systematiccomparisonoftaskadaptationtechniquesfordigitalhistopathology AT georglodde systematiccomparisonoftaskadaptationtechniquesfordigitalhistopathology AT felixnensa systematiccomparisonoftaskadaptationtechniquesfordigitalhistopathology AT dirkschadendorf systematiccomparisonoftaskadaptationtechniquesfordigitalhistopathology AT elisabethlivingstone systematiccomparisonoftaskadaptationtechniquesfordigitalhistopathology AT markuskukuk systematiccomparisonoftaskadaptationtechniquesfordigitalhistopathology |