IMPLEMENTATION OF A COMMERCIAL MODEL OF ARTIFICIAL INTELLIGENCE FOR KARYOTYPING
Aims: Chromosomal banding analysis is the standard technique to identify cytogenetic abnormalities in both constitutional chromosomal disorders and hematological malignancies. Karyotyping is a laborious and manual technique. The development of image (metaphases) capture systems and software for kary...
| Main Authors: | , , , , , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Elsevier
2023-10-01
|
| Series: | Hematology, Transfusion and Cell Therapy |
| Online Access: | http://www.sciencedirect.com/science/article/pii/S2531137923016863 |
| _version_ | 1797654737590419456 |
|---|---|
| author | D Borri RK Kishimoto MFMD Santos RO Safranauskas MG Cordeiro AAC Coimbra GSE Silva JL Silva E Velloso JG Silva |
| author_facet | D Borri RK Kishimoto MFMD Santos RO Safranauskas MG Cordeiro AAC Coimbra GSE Silva JL Silva E Velloso JG Silva |
| author_sort | D Borri |
| collection | DOAJ |
| description | Aims: Chromosomal banding analysis is the standard technique to identify cytogenetic abnormalities in both constitutional chromosomal disorders and hematological malignancies. Karyotyping is a laborious and manual technique. The development of image (metaphases) capture systems and software for karyotyping had a great impact on the routine of cytogenetic laboratories, and recently artificial intelligence (deep neural networks, DNN) has helped in the correct segmentation and chromosome classification, allowing the performance of karyotypes more quickly, with final supervision by experienced cytogeneticists. Some systems using AI are commercially available and our objective was to validate and implement one of these models in our laboratory and measure its productivity. Materials and methods: IA DNN –PC/GTX 1650 version 1.1.40 was configured in the automated metaphase image scanning equipment (Metafer), which uses the software for image analysis (Ikaros), all developed by Metasystems (Altlussheim, Germany). Sixteen machine learning models (classifiers) were tested for bone marrow (oncohematological karyotype) and another 16 for peripheral blood (constitutional karyotype), eight of them with the function of chromosome segmentation in metaphase (separation) and 8 of them with the function of classify chromosomes (chromosomal pairing). 55 metaphases from routine studies of bone marrow for hematological malignancies [300-400 bands, 8 with numerical (−Y, −7) or structural (20q−) chromosomal alterations] and 65 metaphases from PHA-stimulated peripheral blood [550 bands, 7 cases with multiple overlapping chromosomes and 9 with numerical (−X, +13, +21) chromosomal alterations]. The time for manual and automated analysis (time for segmentation + classification + review) was compared. The chromosomal mispairing was noted. Productivity was evaluated (percentage difference in seconds manual-automated analysis/ manual analysis in seconds ×100). Results: The average time for manual analysis of one metaphase in bone marrow samples was 74 ± 30 seconds (segmentation 34 seconds and classification 40 seconds) and in DNN it was 23 ± 19 seconds (segmentation 13 seconds, classification 10 seconds), with a gain of 51 ± 26 seconds and productivity gain of 68.6%, and mismatch of 1 ± 1.5 chromosomes. The average time for manual analysis of one metaphase in peripheral blood samples was 138 ± 56 seconds (segmentation 85 seconds and classification 52 seconds) and in DNN it was 29 ± 16 seconds (segmentation 15 seconds, classification 14 seconds), with a gain of 108 ± 49 seconds and productivity gain of 78.7%, and pairing error of 2.6 ± 2. Discussion: AI allowed a significantly gain in productivity and TAT reduction in karyotype, the constitutional from 21 days to 15 days and the oncohematological karyotype from 14 days to 7 days (5 days for acute leukemias at diagnosis). Conclusion: The use of AI in karyotype analysis proved to be effective and efficient in our laboratory, even using a commercial image bank (DNN). There was an important gain in productivity and the possibility of reducing the TAT, adapting to international quality standards. |
| first_indexed | 2024-03-11T17:03:51Z |
| format | Article |
| id | doaj.art-0121da7238024b0498dc1c0935d58652 |
| institution | Directory Open Access Journal |
| issn | 2531-1379 |
| language | English |
| last_indexed | 2024-03-11T17:03:51Z |
| publishDate | 2023-10-01 |
| publisher | Elsevier |
| record_format | Article |
| series | Hematology, Transfusion and Cell Therapy |
| spelling | doaj.art-0121da7238024b0498dc1c0935d586522023-10-20T06:46:54ZengElsevierHematology, Transfusion and Cell Therapy2531-13792023-10-0145S837S838IMPLEMENTATION OF A COMMERCIAL MODEL OF ARTIFICIAL INTELLIGENCE FOR KARYOTYPINGD Borri0RK Kishimoto1MFMD Santos2RO Safranauskas3MG Cordeiro4AAC Coimbra5GSE Silva6JL Silva7E Velloso8JG Silva9Sociedade Beneficente Israelita Brasileira Albert Einstein (SBIBAE), São Paulo, SP, BrazilSociedade Beneficente Israelita Brasileira Albert Einstein (SBIBAE), São Paulo, SP, BrazilSociedade Beneficente Israelita Brasileira Albert Einstein (SBIBAE), São Paulo, SP, BrazilSociedade Beneficente Israelita Brasileira Albert Einstein (SBIBAE), São Paulo, SP, BrazilSociedade Beneficente Israelita Brasileira Albert Einstein (SBIBAE), São Paulo, SP, BrazilSociedade Beneficente Israelita Brasileira Albert Einstein (SBIBAE), São Paulo, SP, BrazilSociedade Beneficente Israelita Brasileira Albert Einstein (SBIBAE), São Paulo, SP, BrazilSociedade Beneficente Israelita Brasileira Albert Einstein (SBIBAE), São Paulo, SP, BrazilSociedade Beneficente Israelita Brasileira Albert Einstein (SBIBAE), São Paulo, SP, Brazil; Hospital das Clínicas da Faculdade de Medicina da Universidade de São Paulo (HC-FMUSP), São Paulo, SP, BrazilSociedade Beneficente Israelita Brasileira Albert Einstein (SBIBAE), São Paulo, SP, BrazilAims: Chromosomal banding analysis is the standard technique to identify cytogenetic abnormalities in both constitutional chromosomal disorders and hematological malignancies. Karyotyping is a laborious and manual technique. The development of image (metaphases) capture systems and software for karyotyping had a great impact on the routine of cytogenetic laboratories, and recently artificial intelligence (deep neural networks, DNN) has helped in the correct segmentation and chromosome classification, allowing the performance of karyotypes more quickly, with final supervision by experienced cytogeneticists. Some systems using AI are commercially available and our objective was to validate and implement one of these models in our laboratory and measure its productivity. Materials and methods: IA DNN –PC/GTX 1650 version 1.1.40 was configured in the automated metaphase image scanning equipment (Metafer), which uses the software for image analysis (Ikaros), all developed by Metasystems (Altlussheim, Germany). Sixteen machine learning models (classifiers) were tested for bone marrow (oncohematological karyotype) and another 16 for peripheral blood (constitutional karyotype), eight of them with the function of chromosome segmentation in metaphase (separation) and 8 of them with the function of classify chromosomes (chromosomal pairing). 55 metaphases from routine studies of bone marrow for hematological malignancies [300-400 bands, 8 with numerical (−Y, −7) or structural (20q−) chromosomal alterations] and 65 metaphases from PHA-stimulated peripheral blood [550 bands, 7 cases with multiple overlapping chromosomes and 9 with numerical (−X, +13, +21) chromosomal alterations]. The time for manual and automated analysis (time for segmentation + classification + review) was compared. The chromosomal mispairing was noted. Productivity was evaluated (percentage difference in seconds manual-automated analysis/ manual analysis in seconds ×100). Results: The average time for manual analysis of one metaphase in bone marrow samples was 74 ± 30 seconds (segmentation 34 seconds and classification 40 seconds) and in DNN it was 23 ± 19 seconds (segmentation 13 seconds, classification 10 seconds), with a gain of 51 ± 26 seconds and productivity gain of 68.6%, and mismatch of 1 ± 1.5 chromosomes. The average time for manual analysis of one metaphase in peripheral blood samples was 138 ± 56 seconds (segmentation 85 seconds and classification 52 seconds) and in DNN it was 29 ± 16 seconds (segmentation 15 seconds, classification 14 seconds), with a gain of 108 ± 49 seconds and productivity gain of 78.7%, and pairing error of 2.6 ± 2. Discussion: AI allowed a significantly gain in productivity and TAT reduction in karyotype, the constitutional from 21 days to 15 days and the oncohematological karyotype from 14 days to 7 days (5 days for acute leukemias at diagnosis). Conclusion: The use of AI in karyotype analysis proved to be effective and efficient in our laboratory, even using a commercial image bank (DNN). There was an important gain in productivity and the possibility of reducing the TAT, adapting to international quality standards.http://www.sciencedirect.com/science/article/pii/S2531137923016863 |
| spellingShingle | D Borri RK Kishimoto MFMD Santos RO Safranauskas MG Cordeiro AAC Coimbra GSE Silva JL Silva E Velloso JG Silva IMPLEMENTATION OF A COMMERCIAL MODEL OF ARTIFICIAL INTELLIGENCE FOR KARYOTYPING Hematology, Transfusion and Cell Therapy |
| title | IMPLEMENTATION OF A COMMERCIAL MODEL OF ARTIFICIAL INTELLIGENCE FOR KARYOTYPING |
| title_full | IMPLEMENTATION OF A COMMERCIAL MODEL OF ARTIFICIAL INTELLIGENCE FOR KARYOTYPING |
| title_fullStr | IMPLEMENTATION OF A COMMERCIAL MODEL OF ARTIFICIAL INTELLIGENCE FOR KARYOTYPING |
| title_full_unstemmed | IMPLEMENTATION OF A COMMERCIAL MODEL OF ARTIFICIAL INTELLIGENCE FOR KARYOTYPING |
| title_short | IMPLEMENTATION OF A COMMERCIAL MODEL OF ARTIFICIAL INTELLIGENCE FOR KARYOTYPING |
| title_sort | implementation of a commercial model of artificial intelligence for karyotyping |
| url | http://www.sciencedirect.com/science/article/pii/S2531137923016863 |
| work_keys_str_mv | AT dborri implementationofacommercialmodelofartificialintelligenceforkaryotyping AT rkkishimoto implementationofacommercialmodelofartificialintelligenceforkaryotyping AT mfmdsantos implementationofacommercialmodelofartificialintelligenceforkaryotyping AT rosafranauskas implementationofacommercialmodelofartificialintelligenceforkaryotyping AT mgcordeiro implementationofacommercialmodelofartificialintelligenceforkaryotyping AT aaccoimbra implementationofacommercialmodelofartificialintelligenceforkaryotyping AT gsesilva implementationofacommercialmodelofartificialintelligenceforkaryotyping AT jlsilva implementationofacommercialmodelofartificialintelligenceforkaryotyping AT evelloso implementationofacommercialmodelofartificialintelligenceforkaryotyping AT jgsilva implementationofacommercialmodelofartificialintelligenceforkaryotyping |