{"about":{"site":"https://codewithpapers.app","non_affiliation":"Code with Papers and Syntology are not affiliated with, endorsed by, or sponsored by Papers with Code, Meta, or the pwc-archive mirror.","licence":"CC BY-SA 4.0","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","attribution":"https://codewithpapers.app/attribution","modified":"archive material modified by Syntology; see the attribution page"},"url":"/paper/tversky-loss-function-for-image-segmentation","title":"Tversky loss function for image segmentation using 3D fully convolutional deep networks","arxiv_id":"1706.05721","date":"2017-06-18","proceeding":null,"authors":["Seyed Sadegh Mohseni Salehi","Deniz Erdogmus","Ali Gholipour"],"abstract":"Fully convolutional deep neural networks carry out excellent potential for\nfast and accurate image segmentation. One of the main challenges in training\nthese networks is data imbalance, which is particularly problematic in medical\nimaging applications such as lesion segmentation where the number of lesion\nvoxels is often much lower than the number of non-lesion voxels. Training with\nunbalanced data can lead to predictions that are severely biased towards high\nprecision but low recall (sensitivity), which is undesired especially in\nmedical applications where false negatives are much less tolerable than false\npositives. Several methods have been proposed to deal with this problem\nincluding balanced sampling, two step training, sample re-weighting, and\nsimilarity loss functions. In this paper, we propose a generalized loss\nfunction based on the Tversky index to address the issue of data imbalance and\nachieve much better trade-off between precision and recall in training 3D fully\nconvolutional deep neural networks. Experimental results in multiple sclerosis\nlesion segmentation on magnetic resonance images show improved F2 score, Dice\ncoefficient, and the area under the precision-recall curve in test data. Based\non these results we suggest Tversky loss function as a generalized framework to\neffectively train deep neural networks.","url_abs":"http://arxiv.org/abs/1706.05721v1","url_pdf":"http://arxiv.org/pdf/1706.05721v1.pdf","source":{"archive":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","row_kind":"abstracts"},"code_links":[{"paper_slug":"tversky-loss-function-for-image-segmentation","repo_url":"https://github.com/SahinTiryaki/Brain-tumor-segmentation-Vgg19UNet","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"tf","reach":null},{"paper_slug":"tversky-loss-function-for-image-segmentation","repo_url":"https://github.com/umd-fire-coml/2020-Object-Detection-In-Aerial-Images","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"tf","reach":{"status":"unanswered"}}],"tasks":[{"task_slug":"image-segmentation","task_name":"Image Segmentation"},{"task_slug":"lesion-segmentation","task_name":"Lesion Segmentation"},{"task_slug":"segmentation","task_name":"Segmentation"},{"task_slug":"semantic-segmentation","task_name":"Semantic Segmentation"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1706.05721","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}