NCI Office of Data Sharing (ODS) Data Jamboree (Abstract Submissions): Submission #15

serial: '15'
sid: '144987'
uuid: 456ee7d5-0746-46e8-b730-1bfd38dea462
uri: /nci/ods-data-jamboree/abstractsubmissions
created: '1750274999'
completed: '1750275016'
changed: '1750275016'
in_draft: '0'
current_page: ''
remote_addr: 10.208.28.250
uid: '0'
langcode: en
webform_id: nci_office_of_data_sharing_abstr
entity_type: node
entity_id: '2107'
locked: '0'
sticky: '0'
notes: ''
metatag: meta
data:
  category: 'Development or refinement of analysis pipelines or AI/ML algorithms'
  degree_s_: Ph.D.
  email: minkyu.park@nih.gov
  first_name: Minkyu
  keywords_abstracts: 'Risk stratification, Whole Slide Images, Deep Learning, Clinical data'
  last_name: Park
  middle_initial: ''
  organization: 'Computational Genomics & Bioinformatics/Center for Biomedical Informatics & Information Technology/National Cancer Institute'
  organization_address:
    address: ''
    address_2: ''
    city: Rockville
    country: ''
    postal_code: ''
    state_province: ''
  other_please_specify_: ''
  summary: 'Effective risk stratification of cancer patients is critical for precision oncology, particularly when leveraging deep learning models that integrate whole-slide images (WSIs) with omics and clinical data. Although sparse data collection has traditionally hindered the development of robust models, the increasing availability of curated datasets now offers significant opportunities for improvement. In response to this need, we propose an evolutionary, web-based deep learning platform for cancer risk stratification that encompasses an end-to-end training pipeline alongside a continuous inference pipeline for model evaluation. Building on our previous work, we have developed a training pipeline that leverages WSIs and omics data to construct risk stratification models. The platform will be initially deployed using existing childhood cancer data. As new data becomes available, the system automatically integrates this information, updates performance metrics via the inference pipeline, and retrains the model using the training pipeline. This iterative process promotes the gradual evolution and enhancement of model accuracy, with performance changes monitored at each update cycle. The platform will be applicable to all types of childhood cancer, and the new datasets implemented with this platform will enable the display of the status of deep learning model training using the current datasets. Once the model achieves robust performance, it can be employed for real-world predictions, thereby significantly enhancing the utility of the underlying datasets in clinical decision-making.'
  title: 'Senior Bioinformatician'
  ttile: 'An evolutionary deep learning platform for risk stratification in cancer patients.'
  upload_abstract: '65542'