Distance-weighted Sinkhorn loss for Alzheimer’s disease classification
Summary: Traditional loss functions such as cross-entropy loss often quantify the penalty for each mis-classified training sample without adequately considering its distance from the ground truth class distribution in the feature space. Intuitively, the larger this distance is, the higher the penalt...
| Main Authors: | , , , , , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Elsevier
2024-03-01
|
| Series: | iScience |
| Subjects: | |
| Online Access: | http://www.sciencedirect.com/science/article/pii/S2589004224004334 |
| _version_ | 1797296185648611328 |
|---|---|
| author | Zexuan Wang Qipeng Zhan Boning Tong Shu Yang Bojian Hou Heng Huang Andrew J. Saykin Paul M. Thompson Christos Davatzikos Li Shen |
| author_facet | Zexuan Wang Qipeng Zhan Boning Tong Shu Yang Bojian Hou Heng Huang Andrew J. Saykin Paul M. Thompson Christos Davatzikos Li Shen |
| author_sort | Zexuan Wang |
| collection | DOAJ |
| description | Summary: Traditional loss functions such as cross-entropy loss often quantify the penalty for each mis-classified training sample without adequately considering its distance from the ground truth class distribution in the feature space. Intuitively, the larger this distance is, the higher the penalty should be. With this observation, we propose a penalty called distance-weighted Sinkhorn (DWS) loss. For each mis-classified training sample (with predicted label A and true label B), its contribution to the DWS loss positively correlates to the distance the training sample needs to travel to reach the ground truth distribution of all the A samples. We apply the DWS framework with a neural network to classify different stages of Alzheimer’s disease. Our empirical results demonstrate that the DWS framework outperforms the traditional neural network loss functions and is comparable or better to traditional machine learning methods, highlighting its potential in biomedical informatics and data science. |
| first_indexed | 2024-03-07T21:59:43Z |
| format | Article |
| id | doaj.art-0a287bbe671647cb960f6912d4ffa986 |
| institution | Directory Open Access Journal |
| issn | 2589-0042 |
| language | English |
| last_indexed | 2024-03-07T21:59:43Z |
| publishDate | 2024-03-01 |
| publisher | Elsevier |
| record_format | Article |
| series | iScience |
| spelling | doaj.art-0a287bbe671647cb960f6912d4ffa9862024-02-24T04:55:17ZengElsevieriScience2589-00422024-03-01273109212Distance-weighted Sinkhorn loss for Alzheimer’s disease classificationZexuan Wang0Qipeng Zhan1Boning Tong2Shu Yang3Bojian Hou4Heng Huang5Andrew J. Saykin6Paul M. Thompson7Christos Davatzikos8Li Shen9University of Pennsylvania, B301 Richards Building, 3700 Hamilton Walk, Philadelphia, PA 19104, USAUniversity of Pennsylvania, B301 Richards Building, 3700 Hamilton Walk, Philadelphia, PA 19104, USAUniversity of Pennsylvania, B301 Richards Building, 3700 Hamilton Walk, Philadelphia, PA 19104, USAUniversity of Pennsylvania, B301 Richards Building, 3700 Hamilton Walk, Philadelphia, PA 19104, USAUniversity of Pennsylvania, B301 Richards Building, 3700 Hamilton Walk, Philadelphia, PA 19104, USAUniversity of Maryland, College Park, 8125 Paint Branch Drive, College Park, MD 20742, USAIndiana University, 355 West 16th Street, Indianapolis, IN 46202, USAUniversity of Southern California, 4676 Admiralty Way, Marina Del Rey, CA 90292, USAUniversity of Pennsylvania, B301 Richards Building, 3700 Hamilton Walk, Philadelphia, PA 19104, USAUniversity of Pennsylvania, B301 Richards Building, 3700 Hamilton Walk, Philadelphia, PA 19104, USA; Corresponding authorSummary: Traditional loss functions such as cross-entropy loss often quantify the penalty for each mis-classified training sample without adequately considering its distance from the ground truth class distribution in the feature space. Intuitively, the larger this distance is, the higher the penalty should be. With this observation, we propose a penalty called distance-weighted Sinkhorn (DWS) loss. For each mis-classified training sample (with predicted label A and true label B), its contribution to the DWS loss positively correlates to the distance the training sample needs to travel to reach the ground truth distribution of all the A samples. We apply the DWS framework with a neural network to classify different stages of Alzheimer’s disease. Our empirical results demonstrate that the DWS framework outperforms the traditional neural network loss functions and is comparable or better to traditional machine learning methods, highlighting its potential in biomedical informatics and data science.http://www.sciencedirect.com/science/article/pii/S2589004224004334Medical informaticsBiocomputational methodClassification of bioinformatical subjectNeural networksMachine learning |
| spellingShingle | Zexuan Wang Qipeng Zhan Boning Tong Shu Yang Bojian Hou Heng Huang Andrew J. Saykin Paul M. Thompson Christos Davatzikos Li Shen Distance-weighted Sinkhorn loss for Alzheimer’s disease classification iScience Medical informatics Biocomputational method Classification of bioinformatical subject Neural networks Machine learning |
| title | Distance-weighted Sinkhorn loss for Alzheimer’s disease classification |
| title_full | Distance-weighted Sinkhorn loss for Alzheimer’s disease classification |
| title_fullStr | Distance-weighted Sinkhorn loss for Alzheimer’s disease classification |
| title_full_unstemmed | Distance-weighted Sinkhorn loss for Alzheimer’s disease classification |
| title_short | Distance-weighted Sinkhorn loss for Alzheimer’s disease classification |
| title_sort | distance weighted sinkhorn loss for alzheimer s disease classification |
| topic | Medical informatics Biocomputational method Classification of bioinformatical subject Neural networks Machine learning |
| url | http://www.sciencedirect.com/science/article/pii/S2589004224004334 |
| work_keys_str_mv | AT zexuanwang distanceweightedsinkhornlossforalzheimersdiseaseclassification AT qipengzhan distanceweightedsinkhornlossforalzheimersdiseaseclassification AT boningtong distanceweightedsinkhornlossforalzheimersdiseaseclassification AT shuyang distanceweightedsinkhornlossforalzheimersdiseaseclassification AT bojianhou distanceweightedsinkhornlossforalzheimersdiseaseclassification AT henghuang distanceweightedsinkhornlossforalzheimersdiseaseclassification AT andrewjsaykin distanceweightedsinkhornlossforalzheimersdiseaseclassification AT paulmthompson distanceweightedsinkhornlossforalzheimersdiseaseclassification AT christosdavatzikos distanceweightedsinkhornlossforalzheimersdiseaseclassification AT lishen distanceweightedsinkhornlossforalzheimersdiseaseclassification |