Automatic Multi-Label ECG Classification with Category Imbalance and Cost-Sensitive Thresholding
AuthorsLiu, Yang; orcid: 0000-0002-0857-9905; email: 16B903062@stu.hit.edu.cn
Li, Qince; email: email@example.com
Wang, Kuanquan; orcid: 0000-0003-1347-3491; email: firstname.lastname@example.org
Liu, Jun; email: 20B903091@stu.hit.edu.cn
He, Runnan; email: email@example.com
Yuan, Yongfeng; email: firstname.lastname@example.org
Zhang, Henggui; email: email@example.com
MetadataShow full item record
AbstractAutomatic electrocardiogram (ECG) classification is a promising technology for the early screening and follow-up management of cardiovascular diseases. It is, by nature, a multi-label classification task owing to the coexistence of different kinds of diseases, and is challenging due to the large number of possible label combinations and the imbalance among categories. Furthermore, the task of multi-label ECG classification is cost-sensitive, a fact that has usually been ignored in previous studies on the development of the model. To address these problems, in this work, we propose a novel deep learning model–based learning framework and a thresholding method, namely category imbalance and cost-sensitive thresholding (CICST), to incorporate prior knowledge about classification costs and the characteristic of category imbalance in designing a multi-label ECG classifier. The learning framework combines a residual convolutional network with a class-wise attention mechanism. We evaluate our method with a cost-sensitive metric on multiple realistic datasets. The results show that CICST achieved a cost-sensitive metric score of 0.641 ± 0.009 in a 5-fold cross-validation, outperforming other commonly used thresholding methods, including rank-based thresholding, proportion-based thresholding, and fixed thresholding. This demonstrates that, by taking into account the category imbalance and predefined cost information, our approach is effective in improving the performance and practicability of multi-label ECG classification models.
CitationBiosensors, volume 11, issue 11, page e453
DescriptionFrom MDPI via Jisc Publications Router
History: accepted 2021-11-12, pub-electronic 2021-11-14
Publication status: Published
Funder: Collaborative Innovation Center for Prevention and Treatment of Cardiovascular Disease of Si-chuan Province (CICPTCDSP); Grant(s): xtcx2019-01
Showing items related by title, author, creator and subject.
Gene-Environment Interactions Relevant to Estrogen and Risk of Breast Cancer: Can Gene-Environment Interactions Be Detected Only among Candidate SNPs from Genome-Wide Association Studies?Park, JooYong; email: firstname.lastname@example.org; Choi, Ji-Yeob; email: email@example.com; Choi, Jaesung; email: firstname.lastname@example.org; Chung, Seokang; email: email@example.com; Song, Nan; orcid: 0000-0002-9182-1060; email: firstname.lastname@example.org; Park, Sue K.; orcid: 0000-0001-5002-9707; email: email@example.com; Han, Wonshik; email: firstname.lastname@example.org; Noh, Dong-Young; email: email@example.com; Ahn, Sei-Hyun; email: firstname.lastname@example.org; Lee, Jong Won; email: email@example.com; et al. (MDPI, 2021-05-14)In this study we aim to examine gene–environment interactions (GxEs) between genes involved with estrogen metabolism and environmental factors related to estrogen exposure. GxE analyses were conducted with 1970 Korean breast cancer cases and 2052 controls in the case-control study, the Seoul Breast Cancer Study (SEBCS). A total of 11,555 SNPs from the 137 candidate genes were included in the GxE analyses with eight established environmental factors. A replication test was conducted by using an independent population from the Breast Cancer Association Consortium (BCAC), with 62,485 Europeans and 9047 Asians. The GxE tests were performed by using two-step methods in GxEScan software. Two interactions were found in the SEBCS. The first interaction was shown between rs13035764 of NCOA1 and age at menarche in the GE|2df model (p-2df = 1.2 × 10−3). The age at menarche before 14 years old was associated with the high risk of breast cancer, and the risk was higher when subjects had homozygous minor allele G. The second GxE was shown between rs851998 near ESR1 and height in the GE|2df model (p-2df = 1.1 × 10−4). Height taller than 160 cm was associated with a high risk of breast cancer, and the risk increased when the minor allele was added. The findings were not replicated in the BCAC. These results would suggest specificity in Koreans for breast cancer risk.
No Difference in Penetrance between Truncating and Missense/Aberrant Splicing Pathogenic Variants in MLH1 and MSH2: A Prospective Lynch Syndrome Database StudyDominguez-Valentin, Mev; orcid: 0000-0001-7856-0057; email: Mev.Dominguez.Valentin@rr-research.no; Plazzer, John-Paul; orcid: 0000-0001-5114-4301; email: firstname.lastname@example.org; Sampson, Julian R.; email: Sampson@cardiff.ac.uk; Engel, Christoph; orcid: 0000-0002-7247-282X; email: email@example.com; Aretz, Stefan; orcid: 0000-0002-5228-1890; email: firstname.lastname@example.org; Jenkins, Mark A.; email: email@example.com; Sunde, Lone; email: firstname.lastname@example.org; Bernstein, Inge; email: email@example.com; Capella, Gabriel; orcid: 0000-0002-4669-7320; email: firstname.lastname@example.org; Balaguer, Francesc; orcid: 0000-0002-0206-0539; email: email@example.com; et al. (MDPI, 2021-06-28)Background. Lynch syndrome is the most common genetic predisposition for hereditary cancer. Carriers of pathogenic changes in mismatch repair (MMR) genes have an increased risk of developing colorectal (CRC), endometrial, ovarian, urinary tract, prostate, and other cancers, depending on which gene is malfunctioning. In Lynch syndrome, differences in cancer incidence (penetrance) according to the gene involved have led to the stratification of cancer surveillance. By contrast, any differences in penetrance determined by the type of pathogenic variant remain unknown. Objective. To determine cumulative incidences of cancer in carriers of truncating and missense or aberrant splicing pathogenic variants of the MLH1 and MSH2 genes. Methods. Carriers of pathogenic variants of MLH1 (path_MLH1) and MSH2 (path_MSH2) genes filed in the Prospective Lynch Syndrome Database (PLSD) were categorized as truncating or missense/aberrant splicing according to the InSiGHT criteria for pathogenicity. Results. Among 5199 carriers, 1045 had missense or aberrant splicing variants, and 3930 had truncating variants. Prospective observation years for the two groups were 8205 and 34,141 years, respectively, after which there were no significant differences in incidences for cancer overall or for colorectal cancer or endometrial cancers separately. Conclusion. Truncating and missense or aberrant splicing pathogenic variants were associated with similar average cumulative incidences of cancer in carriers of path MLH1 and path_MSH2.
Development of Randomized Trials in Adults with Medulloblastoma—The Example of EORTC 1634-BTG/NOA-23Hau, Peter; orcid: 0000-0003-3894-5053; email: Peter.Hau@ukr.de; Frappaz, Didier; email: Didier.Frappaz@ihope.fr; Hovey, Elizabeth; email: Elizabeth.Hovey@health.nsw.gov.au; McCabe, Martin G.; orcid: 0000-0002-5138-0707; email: Martin.McCabe@manchester.ac.uk; Pajtler, Kristian W.; email: K.Pajtler@kitz-heidelberg.de; Wiestler, Benedikt; orcid: 0000-0002-2963-7772; email: B.Wiestler@tum.de; Seidel, Clemens; email: Clemens.Seidel@medizin.uni-leipzig.de; Combs, Stephanie E.; email: Stephanie.Combs@tum.de; Dirven, Linda; email: L.Dirven@lumc.nl; Klein, Martin; email: M.Klein@amsterdamumc.nl; et al. (MDPI, 2021-07-09)Medulloblastoma is a rare brain malignancy. Patients after puberty are rare and bear an intermediate prognosis. Standard treatment consists of maximal resection plus radio-chemotherapy. Treatment toxicity is high and produces disabling long-term side effects. The sonic hedgehog (SHH) subgroup is highly overrepresented in the post-pubertal and adult population and can be targeted by smoothened (SMO) inhibitors. No practice-changing prospective randomized data have been generated in adults. The EORTC 1634-BTG/NOA-23 trial will randomize patients between standard-dose vs. reduced-dosed craniospinal radiotherapy and SHH-subgroup patients between the SMO inhibitor sonidegib (OdomzoTM, Sun Pharmaceuticals Industries, Inc., New York, USA) in addition to standard radio-chemotherapy vs. standard radio-chemotherapy alone to improve outcomes in view of decreased radiotherapy-related toxicity and increased efficacy. We will further investigate tumor tissue, blood, and cerebrospinal fluid as well as magnetic resonance imaging and radiotherapy plans to generate information that helps to further improve treatment outcomes. Given that treatment side effects typically occur late, long-term follow-up will monitor classic side effects of therapy, but also health-related quality of life, cognition, social and professional outcome, and reproduction and fertility. In summary, we will generate unprecedented data that will be translated into treatment changes in post-pubertal patients with medulloblastoma and will help to design future clinical trials.