66216e109a540ff2e6f07fac8d1d0a07 electronics_a2022v11n9p1447.pdf a0c62bf0f79382eeba4695785fea9a421d37f7c3 electronics_a2022v11n9p1447.pdf 17ec1552ffa381d8d1d3a3ba28c701904a062122da9f19a3e09ac9bb05f906de electronics_a2022v11n9p1447.pdf Title: Transfer Learning Improving Predictive Mortality Models for Patients in End-Stage Renal Disease Subject: Deep learning is becoming a fundamental piece in the paradigm shift from evidence-based to data-based medicine. However, its learning capacity is rarely exploited when working with small data sets. Through transfer learning (TL), information from a source domain is transferred to a target one to enhance a learning task in such domain. The proposed TL mechanisms are based on sample and feature space augmentation. Thus, deep autoencoders extract complex representations for the data in the TL approach. Their latent representations, the so-called codes, are handled to transfer information among domains. The transfer of samples is carried out by computing a latent space mapping matrix that links codes from both domains for later reconstruction. The feature space augmentation is based on the computation of the average of the most similar codes from one domain. Such an average augments the features in a target domain. The proposed framework is evaluated in the prediction of mortality in patients in end-stage renal disease, transferring information related to the mortality of patients with acute kidney injury from the massive database MIMIC-III. Compared to other TL mechanisms, the proposed approach improves 6–11% in previous mortality predictive models. The integration of TL approaches into learning tasks in pathologies with data volume issues could encourage the use of data-based medicine in a clinical setting. Keywords: transfer learning; deep learning; mortality prediction Author: Edwar Macias, Jose Lopez Vicario, Javier Serrano, Jose Ibeas and Antoni Morell Creator: LaTeX with hyperref Producer: pdfTeX-1.40.21 CreationDate: Sat Apr 30 10:34:38 2022 CEST ModDate: Sat Apr 30 10:37:58 2022 CEST Tagged: no UserProperties: no Suspects: no Form: none JavaScript: no Pages: 14 Encrypted: no Page size: 595.276 x 841.89 pts (A4) Page rot: 0 File size: 2509057 bytes Optimized: no PDF version: 1.7 name type encoding emb sub uni object ID ------------------------------------ ----------------- ---------------- --- --- --- --------- JZRSYK+URWPalladioL-Bold Type 1 Custom yes yes no 10 0 VSXWUQ+URWPalladioL-Roma Type 1 Custom yes yes no 15 0 PRUDII+URWPalladioL-Ital Type 1 Custom yes yes no 19 0 GKSXYT+TimesNewRomanPS-BoldItalicMT Type 1C WinAnsi yes yes no 37 0 IJPTNE+CMSY10 Type 1 Builtin yes yes no 59 0 VHYTKV+CMEX10 Type 1 Builtin yes yes no 66 0 QSYHNX+PazoMath-Italic Type 1 Builtin yes yes no 70 0 SOSTRQ+CMR10 Type 1 Builtin yes yes no 74 0 IKXQUG+PazoMath Type 1 Builtin yes yes no 78 0 SPTXKM+HelveticaNeue TrueType MacRoman yes yes no 93 0 LTVNOJ+TimesNewRomanPS-BoldItalicMT TrueType MacRoman yes yes no 97 0 UUBPPN+TimesNewRomanPS-ItalicMT TrueType MacRoman yes yes no 101 0 PYKEPA+TimesNewRomanPS-BoldMT TrueType MacRoman yes yes no 105 0 HNYFFX+TimesNewRomanPSMT TrueType MacRoman yes yes no 109 0 ZSHSFB+STIXGeneral-Regular Type 1 Custom yes yes yes 113 0 XSBJNE+MSBM10 Type 1 Builtin yes yes no 123 0 OYLZXO+HelveticaNeue TrueType MacRoman yes yes no 140 0 BXWQUP+TimesNewRomanPS-BoldItalicMT TrueType MacRoman yes yes no 144 0 HJUCUF+TimesNewRomanPS-ItalicMT TrueType MacRoman yes yes no 148 0 YZAEPE+AmericanTypewriter-Light TrueType WinAnsi yes yes yes 152 0 SDJOQY+TimesNewRomanPS-BoldMT TrueType MacRoman yes yes no 158 0 UELEOM+TimesNewRomanPSMT TrueType MacRoman yes yes no 162 0 ZSHSFB+STIXGeneral-Regular Type 1 Custom yes yes yes 166 0 XTXSKG+HelveticaNeue TrueType MacRoman yes yes no 186 0 SKIQWV+HelveticaNeue-Bold TrueType MacRoman yes yes no 190 0 PPREXB+HelveticaNeue TrueType MacRoman yes yes no 213 0 IRUXKV+HelveticaNeue-Bold TrueType MacRoman yes yes no 217 0 BJPRVF+Times-Italic TrueType MacRoman yes yes no 221 0 AVVYFM+HelveticaNeue TrueType MacRoman yes yes no 241 0 LRSZDT+HelveticaNeue-Bold TrueType MacRoman yes yes no 245 0 LCFKHH+Times-Italic TrueType MacRoman yes yes no 249 0 MUTLQI+HelveticaNeue TrueType MacRoman yes yes no 272 0 KLSZRJ+Times-Italic TrueType MacRoman yes yes no 276 0 JFDXBM+Helvetica-Light TrueType MacRoman yes yes no 280 0 KLCLXW+LucidaGrande TrueType MacRoman yes yes no 284 0 ZSHSFB+STIXGeneral-Regular Type 1 Custom yes yes yes 288 0 SSKLIO+HelveticaNeue TrueType MacRoman yes yes no 314 0 OACMZM+PazoMath-BoldItalic Type 1 Builtin yes yes no 337 0 Jhove (Rel. 1.22.1, 2019-04-17) Date: 2022-05-07 02:28:03 CEST RepresentationInformation: electronics_a2022v11n9p1447.pdf ReportingModule: PDF-hul, Rel. 1.12.1 (2019-04-17) LastModified: 2022-05-06 10:38:16 CEST Size: 2509057 Format: PDF Version: 1.7 Status: Well-Formed and valid SignatureMatches: PDF-hul MIMEtype: application/pdf PDFMetadata: Objects: 476 FreeObjects: 57 IncrementalUpdates: 0 DocumentCatalog: PageLayout: SinglePage PageMode: UseNone Outlines: Item: Title: Introduction Destination: edu.harvard.hul.ois.jhove.module.pdf.PdfSimpleObject@7b69c6ba Item: Title: Materials and Methods Destination: edu.harvard.hul.ois.jhove.module.pdf.PdfSimpleObject@46daef40 Children: Item: Title: Autoencoders Destination: edu.harvard.hul.ois.jhove.module.pdf.PdfSimpleObject@12f41634 Item: Title: Hybrid Heterogeneous Transfer Learning Destination: edu.harvard.hul.ois.jhove.module.pdf.PdfSimpleObject@13c27452 Item: Title: Problem Definition Destination: edu.harvard.hul.ois.jhove.module.pdf.PdfSimpleObject@262b2c86 Item: Title: Proposed Method Destination: edu.harvard.hul.ois.jhove.module.pdf.PdfSimpleObject@371a67ec Children: Item: Title: Sample Augmentation TLCO Destination: edu.harvard.hul.ois.jhove.module.pdf.PdfSimpleObject@5ed828d Item: Title: Feature Space Augmentation TLAV Destination: edu.harvard.hul.ois.jhove.module.pdf.PdfSimpleObject@50d0686 Item: Title: Experimental Setup Destination: edu.harvard.hul.ois.jhove.module.pdf.PdfSimpleObject@7a3d45bd Item: Title: Results Destination: edu.harvard.hul.ois.jhove.module.pdf.PdfSimpleObject@1e7c7811 Children: Item: Title: TLCO Sample Augmentation Destination: edu.harvard.hul.ois.jhove.module.pdf.PdfSimpleObject@77ec78b9 Item: Title: TLAV Feature Space Augmentation Destination: edu.harvard.hul.ois.jhove.module.pdf.PdfSimpleObject@1a3869f4 Item: Title: TLAV HHTLM Destination: edu.harvard.hul.ois.jhove.module.pdf.PdfSimpleObject@a38d7a3 Item: Title: Discussion Destination: edu.harvard.hul.ois.jhove.module.pdf.PdfSimpleObject@77f99a05 Item: Title: References Destination: edu.harvard.hul.ois.jhove.module.pdf.PdfSimpleObject@63440df3 Info: Title: Transfer Learning Improving Predictive Mortality Models for Patients in End-Stage Renal Disease Author: Edwar Macias, Jose Lopez Vicario, Javier Serrano, Jose Ibeas and Antoni Morell Subject: Deep learning is becoming a fundamental piece in the paradigm shift from evidence-based to data-based medicine. However, its learning capacity is rarely exploited when working with small data sets. Through transfer learning (TL), information from a source domain is transferred to a target one to enhance a learning task in such domain. The proposed TL mechanisms are based on sample and feature space augmentation. Thus, deep autoencoders extract complex representations for the data in the TL approach. Their latent representations, the so-called codes, are handled to transfer information among domains. The transfer of samples is carried out by computing a latent space mapping matrix that links codes from both domains for later reconstruction. The feature space augmentation is based on the computation of the average of the most similar codes from one domain. Such an average augments the features in a target domain. The proposed framework is evaluated in the prediction of mortality in patients in end-stage renal disease, transferring information related to the mortality of patients with acute kidney injury from the massive database MIMIC-III. Compared to other TL mechanisms, the proposed approach improves 6 11% in previous mortality predictive models. The integration of TL approaches into learning tasks in pathologies with data volume issues could encourage the use of data-based medicine in a clinical setting. Keywords: transfer learning; deep learning; mortality prediction Creator: LaTeX with hyperref Producer: pdfTeX-1.40.21 CreationDate: Sat Apr 30 10:34:38 CEST 2022 ModDate: Sat Apr 30 10:37:58 CEST 2022 ID: 0xc903ab73fe49010d4f85eb7e993033c8, 0xc903ab73fe49010d4f85eb7e993033c8 Filters: FilterPipeline: FlateDecode Fonts: Type1: Font: BaseFont: OACMZM+PazoMath-BoldItalic FontSubset: true FirstChar: 101 LastChar: 101 FontDescriptor: FontName: OACMZM+PazoMath-BoldItalic Flags: Symbolic FontBBox: -64, -272, 868, 740 FontFile: true Font: BaseFont: VHYTKV+CMEX10 FontSubset: true FirstChar: 0 LastChar: 115 FontDescriptor: FontName: VHYTKV+CMEX10 Flags: Symbolic FontBBox: -24, -2960, 1454, 772 FontFile: true Font: BaseFont: PRUDII+URWPalladioL-Ital FontSubset: true FirstChar: 2 LastChar: 151 FontDescriptor: FontName: PRUDII+URWPalladioL-Ital Flags: Symbolic FontBBox: -170, -305, 1010, 941 FontFile: true EncodingDictionary: Differences: true Font: BaseFont: QSYHNX+PazoMath-Italic FontSubset: true FirstChar: 101 LastChar: 113 FontDescriptor: FontName: QSYHNX+PazoMath-Italic Flags: Symbolic FontBBox: -70, -277, 902, 733 FontFile: true Font: BaseFont: JZRSYK+URWPalladioL-Bold FontSubset: true FirstChar: 3 LastChar: 122 FontDescriptor: FontName: JZRSYK+URWPalladioL-Bold Flags: Symbolic FontBBox: -152, -301, 1000, 935 FontFile: true EncodingDictionary: Differences: true Font: BaseFont: SOSTRQ+CMR10 FontSubset: true FirstChar: 40 LastChar: 93 FontDescriptor: FontName: SOSTRQ+CMR10 Flags: Symbolic FontBBox: -40, -250, 1009, 750 FontFile: true Font: BaseFont: IJPTNE+CMSY10 FontSubset: true FirstChar: 0 LastChar: 107 FontDescriptor: FontName: IJPTNE+CMSY10 Flags: Symbolic FontBBox: -29, -960, 1116, 775 FontFile: true Font: BaseFont: XSBJNE+MSBM10 FontSubset: true FirstChar: 69 LastChar: 82 FontDescriptor: FontName: XSBJNE+MSBM10 Flags: Symbolic FontBBox: -55, -420, 2343, 920 FontFile: true Font: BaseFont: IKXQUG+PazoMath FontSubset: true FirstChar: 165 LastChar: 229 FontDescriptor: FontName: IKXQUG+PazoMath Flags: Symbolic FontBBox: -40, -283, 878, 946 FontFile: true Font: BaseFont: VSXWUQ+URWPalladioL-Roma FontSubset: true FirstChar: 2 LastChar: 243 FontDescriptor: FontName: VSXWUQ+URWPalladioL-Roma Flags: Symbolic FontBBox: -166, -283, 1021, 943 FontFile: true EncodingDictionary: Differences: true Pages: Page: Sequence: 1 Annotations: Annotation: Subtype: Link Rect: 115, 335, 152, 344 Annotation: Subtype: Link Rect: 34, 323, 150, 332 Annotation: Subtype: Link Rect: 34, 164, 96, 186 Annotation: Subtype: Link Rect: 126, 90, 154, 99 Annotation: Subtype: Link Rect: 34, 78, 147, 87 Annotation: Subtype: Link Rect: 34, 756, 170, 792 Annotation: Subtype: Link Rect: 515, 756, 560, 792 Annotation: Subtype: Link Rect: 111, 661, 122, 674 Annotation: Subtype: Link Rect: 216, 661, 227, 674 Annotation: Subtype: Link Rect: 301, 661, 313, 674 Annotation: Subtype: Link Rect: 368, 661, 379, 674 Annotation: Subtype: Link Rect: 472, 661, 483, 674 Annotation: Subtype: Link Rect: 482, 268, 489, 277 ActionDest: 361 Annotation: Subtype: Link Rect: 256, 231, 263, 240 ActionDest: 361 Annotation: Subtype: Link Rect: 549, 205, 556, 215 ActionDest: 361 Annotation: Subtype: Link Rect: 265, 130, 271, 139 ActionDest: 361 Annotation: Subtype: Link Rect: 487, 105, 494, 114 ActionDest: 361 Annotation: Subtype: Link Rect: 497, 105, 504, 114 ActionDest: 361 Annotation: Subtype: Link Rect: 124, 17, 286, 28 Annotation: Subtype: Link Rect: 399, 17, 560, 28 Annotation: Subtype: Link Rect: 34, 408, 87, 427 Page: Sequence: 2 Annotations: Annotation: Subtype: Link Rect: 318, 723, 325, 732 ActionDest: 361 Annotation: Subtype: Link Rect: 328, 723, 340, 732 ActionDest: 361 Annotation: Subtype: Link Rect: 531, 673, 543, 682 ActionDest: 361 Annotation: Subtype: Link Rect: 543, 673, 555, 682 ActionDest: 361 Annotation: Subtype: Link Rect: 493, 572, 505, 581 ActionDest: 361 Annotation: Subtype: Link Rect: 212, 560, 224, 569 ActionDest: 361 Annotation: Subtype: Link Rect: 338, 560, 350, 569 ActionDest: 361 Annotation: Subtype: Link Rect: 355, 547, 367, 556 ActionDest: 361 Annotation: Subtype: Link Rect: 349, 346, 361, 355 ActionDest: 361 Annotation: Subtype: Link Rect: 447, 237, 454, 249 ActionDest: 56 Annotation: Subtype: Link Rect: 400, 225, 407, 237 ActionDest: 292 Annotation: Subtype: Link Rect: 284, 212, 291, 224 ActionDest: 358 Page: Sequence: 3 Annotations: Annotation: Subtype: Link Rect: 274, 667, 281, 679 ActionDest: 63 Annotation: Subtype: Link Rect: 352, 353, 364, 362 ActionDest: 361 Annotation: Subtype: Link Rect: 510, 259, 522, 268 ActionDest: 361 Annotation: Subtype: Link Rect: 338, 246, 350, 255 ActionDest: 361 Annotation: Subtype: Link Rect: 379, 193, 386, 205 ActionDest: 120 Annotation: Subtype: Link Rect: 447, 168, 454, 180 ActionDest: 120 Page: Sequence: 4 Annotations: Annotation: Subtype: Link Rect: 317, 427, 324, 438 ActionDest: 120 Annotation: Subtype: Link Rect: 439, 293, 446, 305 ActionDest: 120 Annotation: Subtype: Link Rect: 535, 135, 547, 144 ActionDest: 361 Page: Sequence: 5 Annotations: Annotation: Subtype: Link Rect: 257, 707, 264, 720 ActionDest: 170 Annotation: Subtype: Link Rect: 521, 695, 528, 707 ActionDest: 170 Annotation: Subtype: Link Rect: 514, 105, 521, 117 ActionDest: 194 Page: Sequence: 6 Annotations: Annotation: Subtype: Link Rect: 466, 108, 473, 120 ActionDest: 225 Annotation: Subtype: Link Rect: 385, 83, 392, 95 ActionDest: 225 Page: Sequence: 7 Annotations: Annotation: Subtype: Link Rect: 485, 141, 493, 153 ActionDest: 253 Annotation: Subtype: Link Rect: 470, 92, 477, 103 ActionDest: 253 Annotation: Subtype: Link Rect: 453, 66, 460, 78 ActionDest: 253 Page: Sequence: 8 Annotations: Annotation: Subtype: Link Rect: 205, 239, 217, 248 ActionDest: 361 Annotation: Subtype: Link Rect: 489, 114, 501, 123 ActionDest: 361 Annotation: Subtype: Link Rect: 327, 76, 339, 85 ActionDest: 361 Page: Sequence: 9 Annotations: Annotation: Subtype: Link Rect: 423, 529, 435, 538 ActionDest: 361 Annotation: Subtype: Link Rect: 222, 489, 236, 501 ActionDest: 120 Annotation: Subtype: Link Rect: 526, 289, 538, 298 ActionDest: 361 Annotation: Subtype: Link Rect: 544, 214, 556, 223 ActionDest: 361 Annotation: Subtype: Link Rect: 190, 198, 197, 210 ActionDest: 295 Page: Sequence: 10 Annotations: Annotation: Subtype: Link Rect: 418, 588, 425, 600 ActionDest: 298 Annotation: Subtype: Link Rect: 213, 337, 220, 349 ActionDest: 298 Annotation: Subtype: Link Rect: 517, 337, 524, 349 ActionDest: 298 Annotation: Subtype: Link Rect: 261, 249, 268, 261 ActionDest: 298 Page: Sequence: 11 Annotations: Annotation: Subtype: Link Rect: 281, 280, 288, 292 ActionDest: 334 Annotation: Subtype: Link Rect: 301, 180, 308, 192 ActionDest: 334 Page: Sequence: 12 Annotations: Annotation: Subtype: Link Rect: 553, 194, 560, 206 ActionDest: 358 Page: Sequence: 13 Annotations: Annotation: Subtype: Link Rect: 165, 80, 306, 91 Page: Sequence: 14 Annotations: Annotation: Subtype: Link Rect: 59, 674, 96, 684 Annotation: Subtype: Link Rect: 102, 674, 138, 684 Annotation: Subtype: Link Rect: 158, 651, 195, 661 Annotation: Subtype: Link Rect: 170, 628, 208, 638 Annotation: Subtype: Link Rect: 456, 616, 493, 627 Annotation: Subtype: Link Rect: 393, 581, 430, 592 Annotation: Subtype: Link Rect: 346, 558, 383, 569 Annotation: Subtype: Link Rect: 413, 535, 451, 546 Annotation: Subtype: Link Rect: 420, 512, 457, 523 Annotation: Subtype: Link Rect: 396, 489, 433, 500 Annotation: Subtype: Link Rect: 59, 455, 96, 465 Annotation: Subtype: Link Rect: 409, 432, 446, 442 Annotation: Subtype: Link Rect: 453, 432, 489, 442 Annotation: Subtype: Link Rect: 479, 408, 515, 419 Annotation: Subtype: Link Rect: 521, 408, 557, 419 Annotation: Subtype: Link Rect: 187, 385, 225, 396 Annotation: Subtype: Link Rect: 231, 385, 267, 396 Annotation: Subtype: Link Rect: 160, 351, 197, 362 Annotation: Subtype: Link Rect: 231, 316, 268, 327 Annotation: Subtype: Link Rect: 272, 293, 310, 304 Annotation: Subtype: Link Rect: 292, 270, 329, 281 Annotation: Subtype: Link Rect: 364, 247, 402, 258 Annotation: Subtype: Link Rect: 59, 225, 96, 235 Annotation: Subtype: Link Rect: 113, 168, 150, 178 Annotation: Subtype: Link Rect: 59, 133, 96, 143 Annotation: Subtype: Link Rect: 102, 133, 138, 143 Checksum: c08da2c2 Type: CRC32 Checksum: 66216e109a540ff2e6f07fac8d1d0a07 Type: MD5 Checksum: a0c62bf0f79382eeba4695785fea9a421d37f7c3 Type: SHA-1