FRINK Linked Data Fragments server

Matches in SemOpenAlex for { <https://semopenalex.org/work/W4385667579> ?p ?o ?g. }

• W4385667579 endingPage "e44236bf" @default.
• W4385667579 startingPage "e44236bf" @default.
• W4385667579 abstract "Topic: 23. Hematopoiesis, stem cells and microenvironment Background: Bone marrow morphology remains a standard diagnostic procedure in any specialised haematology diagnostic laboratory. Cytomorphological evaluation of bone marrow smear consists of two main steps: Counting and evaluating hundreds of bone marrow cells to obtain a myelogram, which is performed by analysts, and secondly identifying the possible underlying hematologic disease, which is performed by haematologists. Both procedures are time and resource consuming. Today, automatic identification of blood cells in blood smears by light microscopy and special image analysis is part of routine blood analysis integrated into automated blood cell counters. However, automatic identification of bone marrow cells is algorithmically more difficult. There are only a handful of in-house academic solutions, usually focused on a specific diagnostic problem, and only one commercial solution suitable for daily clinical use. Aims: We decided to develop our own in-house bone marrow diagnostic environment using artificial intelligence (AI) algorithms to expedite the (1) diagnostic procedure and (2) release human resources. First, we focused on automated report of myelogram which would be followed by producing diagnosis of hematologic condition based on smear analysis. Methods: Fully automized Olympus BX63 microscope guided by Genasi HiPath Pro software was used for automatic scanning of smears stained by Wright-Giemsa and the obtained pictures were collected. User interface for the Annotation Tool was developed in JavaScript. Backend solutions, classification and training were developed in Python using Pytorch, sklearn, scipy libraries. We developed two active models. ResNet as a classification predictor combined with YOLO that does segmentation of images for separate cells. Annotations contain boundaries of cells and can be masked when boundaries overlap in areas. Masks are used to perform per pixel segmentation in pre-training steps. Dataset training was run on Nvidia HGX with 960 GB RAM and 6x Nvidia A100 with 480 GB VRAM. Results: Classification duration per image was 0.022s to 0.095s. Whole-slide processing of 12.000 images lasted from 240 s to 960 s on the same hardware as training was performed. Timing included loading time of image into memory, segmentation, and classification step. Timings when using API calls as a service depended vastly on network speed and image size. To simplify this, we developed a scan uploader where users could upload a whole-slide image which was up to 1 TB and let it run overnight. After upload was complete a segmentation-classification was performed on the machine and results were made available (Table 1). Expected classification rate was 90-97%. Bone marrow smear with cells and the annotation tool can be seen in the Picture 1. Picture 1: Segmentation-classification results, bone marrow smear with detected cells and the annotation tool.With the use of AI we could improve standard bone marrow diagnostics: 1.By scanning not only the cell sample but the entire bone marrow smear. 2.By recording cell positions relative to other cell-types in their neighbourhood. 3.AI identified picture of any cell can be used later to speed up the diagnostic process. Summary/Conclusion: Development of in-house bone marrow morphology AI model based on YOLO/ResNet is feasible. It was successfully trained to identify cells in the pictures of bone marrow smears reducing diagnostic time with expected classification rate 90-97%. With future development it has capabilities to shorten the whole diagnostic process and release human resources for other work in the laboratory. Keywords: Artificial intelligence, Bone Marrow, Machine learning, Diagnosis" @default.
• W4385667579 created "2023-08-09" @default.
• W4385667579 creator A5004340496 @default.
• W4385667579 creator A5012645409 @default.
• W4385667579 creator A5039907428 @default.
• W4385667579 creator A5050815713 @default.
• W4385667579 creator A5092614102 @default.
• W4385667579 creator A5092614103 @default.
• W4385667579 date "2023-08-01" @default.
• W4385667579 modified "2023-09-27" @default.
• W4385667579 title "P1352: BONE MARROW MORPHOLOGY DIAGNOSTIC ENVIRONMENT USING YOLO/RESNET ARTIFICIAL INTELLIGENCE MODEL" @default.
• W4385667579 doi "https://doi.org/10.1097/01.hs9.0000972296.44236.bf" @default.
• W4385667579 hasPublicationYear "2023" @default.
• W4385667579 type Work @default.
• W4385667579 citedByCount "0" @default.
• W4385667579 crossrefType "journal-article" @default.
• W4385667579 hasAuthorship W4385667579A5004340496 @default.
• W4385667579 hasAuthorship W4385667579A5012645409 @default.
• W4385667579 hasAuthorship W4385667579A5039907428 @default.
• W4385667579 hasAuthorship W4385667579A5050815713 @default.
• W4385667579 hasAuthorship W4385667579A5092614102 @default.
• W4385667579 hasAuthorship W4385667579A5092614103 @default.
• W4385667579 hasBestOaLocation W43856675791 @default.
• W4385667579 hasConcept C142724271 @default.
• W4385667579 hasConcept C154945302 @default.
• W4385667579 hasConcept C2780007613 @default.
• W4385667579 hasConcept C41008148 @default.
• W4385667579 hasConcept C71924100 @default.
• W4385667579 hasConceptScore W4385667579C142724271 @default.
• W4385667579 hasConceptScore W4385667579C154945302 @default.
• W4385667579 hasConceptScore W4385667579C2780007613 @default.
• W4385667579 hasConceptScore W4385667579C41008148 @default.
• W4385667579 hasConceptScore W4385667579C71924100 @default.
• W4385667579 hasIssue "S3" @default.
• W4385667579 hasLocation W43856675791 @default.
• W4385667579 hasLocation W43856675792 @default.
• W4385667579 hasOpenAccess W4385667579 @default.
• W4385667579 hasPrimaryLocation W43856675791 @default.
• W4385667579 hasRelatedWork W1970707390 @default.
• W4385667579 hasRelatedWork W2027717767 @default.
• W4385667579 hasRelatedWork W2038751394 @default.
• W4385667579 hasRelatedWork W2067699521 @default.
• W4385667579 hasRelatedWork W2371279320 @default.
• W4385667579 hasRelatedWork W2379364167 @default.
• W4385667579 hasRelatedWork W2748952813 @default.
• W4385667579 hasRelatedWork W2899084033 @default.
• W4385667579 hasRelatedWork W3032513420 @default.
• W4385667579 hasRelatedWork W74504067 @default.
• W4385667579 hasVolume "7" @default.
• W4385667579 isParatext "false" @default.
• W4385667579 isRetracted "false" @default.
• W4385667579 workType "article" @default.