{"id":{"repo_id":"vilnius","oai_identifier":"oai:vu.lt:elaba:308336925"},"canonical_url":"https://search.ndltd.org/etd/vilnius/oai:vu.lt:elaba:308336925","repository":{"repo_id":"vilnius","name":"Vilnius University","base_url":"https://epublications.vu.lt/oai"},"display":{"title":"Design of agentic ai systems for automated data extraction in medical processes /","abstract":"This thesis examines the design and evaluation of an agentic artificial intelligence system for automated data extraction from anonymized Lithuanian pathology reports. The goal is to convert unstructured clinical narrative into structured records that follow International Collaboration on Cancer Reporting (ICCR) datasets while preserving field-level traceability: every extracted value must be linked to a verbatim source span that supports it. To address this task, an agentic extraction system was developed in which an ICCR schema is populated through separate field-level processing steps. The system first retrieves source spans relevant to a field, then extracts a candidate value, verifies it using a Chain-of-Verification step, and emits schema-constrained JSON output. This architecture is compared against a strong one-shot baseline that receives the same case, schema, and field documentation. The evaluation uses a locked suite of 38 reviewed clinical cases covering four ICCR datasets: breast invasive carcinoma, radical prostatectomy, prostate biopsy, and colorectal resection. Both conditions use the same Qwen3 235B backbone model, and the evaluation is performed with the correct ICCR schema supplied by the suite manifest. The agentic system achieves 75.7% micro F1, compared with 63.0% for the one-shot baseline, and improves citation precision from 76.0% to 90.6%. The results show that the agentic control flow improves extraction mainly by reducing false-positive values and semantically unsupported source citations. The main remaining errors involve multi-value fields, numerical interpretation, and the scoring of clinically equivalent free-text paraphrases.","abstract_html":"This thesis examines the design and evaluation of an agentic artificial intelligence system for automated data extraction from anonymized Lithuanian pathology reports. The goal is to convert unstructured clinical narrative into structured records that follow International Collaboration on Cancer Reporting (ICCR) datasets while preserving field-level traceability: every extracted value must be linked to a verbatim source span that supports it. To address this task, an agentic extraction system was developed in which an ICCR schema is populated through separate field-level processing steps. The system first retrieves source spans relevant to a field, then extracts a candidate value, verifies it using a Chain-of-Verification step, and emits schema-constrained JSON output. This architecture is compared against a strong one-shot baseline that receives the same case, schema, and field documentation. The evaluation uses a locked suite of 38 reviewed clinical cases covering four ICCR datasets: breast invasive carcinoma, radical prostatectomy, prostate biopsy, and colorectal resection. Both conditions use the same Qwen3 235B backbone model, and the evaluation is performed with the correct ICCR schema supplied by the suite manifest. The agentic system achieves 75.7% micro F1, compared with 63.0% for the one-shot baseline, and improves citation precision from 76.0% to 90.6%. The results show that the agentic control flow improves extraction mainly by reducing false-positive values and semantically unsupported source citations. The main remaining errors involve multi-value fields, numerical interpretation, and the scoring of clinically equivalent free-text paraphrases.","abstract_has_math":false,"creators":["Grenda, Marijonas,"],"institution":"Institutional Repository of Vilnius University","degree_name":null,"degree_level":null,"degree_discipline":null,"degree_department":null,"school":null,"contributors":[],"advisors":[],"committee_chairs":[],"committee_members":[],"year":2026,"date_issued":"2026","date_published":"2026","updated_at":"2026-11-19T10:12:59Z","subjects":["agentic AI systems, medical information extraction, pathology reports, ICCR, large language models, source traceability, citation precision"],"languages":["eng"],"rights":["info:eu-repo/semantics/openAccess"],"rights_urls":[],"identifier_entries":[],"person_identifier_entries":[]},"links":{"outbound_url":"https://repository.vu.lt/VU:ELABAETD308336925&prefLang=en_US","outbound_label":"Repository record","outbound_source":"dc:identifier","record_url":"https://repository.vu.lt/VU:ELABAETD308336925&prefLang=en_US","record_label":"Repository record","record_kind":"record","record_source":"dc:identifier","download_url":null,"download_label":null,"download_source":null,"access_status":"open","embargo_until":null,"access_note":null},"source_record":{"url":"https://epublications.vu.lt/oai?verb=GetRecord&metadataPrefix=oai_dc&identifier=oai%3Avu.lt%3Aelaba%3A308336925","prefix":"oai_dc"},"metadata_groups":[{"id":"people","label":"People","entries":[{"key":"dc:creator","label":"Author","values":["Grenda, Marijonas,"]}]},{"id":"academic_context","label":"Academic Context","entries":[{"key":"dc:date","label":"Dc Date","values":["2026"]},{"key":"dc:publisher","label":"Institution","values":["Institutional Repository of Vilnius University"]},{"key":"dc:relation","label":"Dc Relation","values":["https://epublications.vu.lt/object/elaba:308336925/308336925.pdf"]},{"key":"dc:type","label":"Dc Type","values":["info:eu-repo/semantics/bachelorThesis"]}]},{"id":"subjects_keywords","label":"Subjects and Keywords","entries":[{"key":"dc:subject","label":"Dc Subject","values":["agentic AI systems, medical information extraction, pathology reports, ICCR, large language models, source traceability, citation precision"]}]},{"id":"language_rights","label":"Language and Rights","entries":[{"key":"dc:language","label":"Dc Language","values":["eng"]},{"key":"dc:rights","label":"Dc Rights","values":["info:eu-repo/semantics/openAccess"]}]},{"id":"identifiers","label":"Identifiers","entries":[{"key":"dc:identifier","label":"Identifier","values":["https://repository.vu.lt/VU:ELABAETD308336925&prefLang=en_US"]}]},{"id":"additional","label":"Additional Metadata","entries":[{"key":"dc:description","label":"Description","values":["This thesis examines the design and evaluation of an agentic artificial intelligence system for automated data extraction from anonymized Lithuanian pathology reports. The goal is to convert unstructured clinical narrative into structured records that follow International Collaboration on Cancer Reporting (ICCR) datasets while preserving field-level traceability: every extracted value must be linked to a verbatim source span that supports it. To address this task, an agentic extraction system was developed in which an ICCR schema is populated through separate field-level processing steps. The system first retrieves source spans relevant to a field, then extracts a candidate value, verifies it using a Chain-of-Verification step, and emits schema-constrained JSON output. This architecture is compared against a strong one-shot baseline that receives the same case, schema, and field documentation. The evaluation uses a locked suite of 38 reviewed clinical cases covering four ICCR datasets: breast invasive carcinoma, radical prostatectomy, prostate biopsy, and colorectal resection. Both conditions use the same Qwen3 235B backbone model, and the evaluation is performed with the correct ICCR schema supplied by the suite manifest. The agentic system achieves 75.7% micro F1, compared with 63.0% for the one-shot baseline, and improves citation precision from 76.0% to 90.6%. The results show that the agentic control flow improves extraction mainly by reducing false-positive values and semantically unsupported source citations. The main remaining errors involve multi-value fields, numerical interpretation, and the scoring of clinically equivalent free-text paraphrases."]},{"key":"dc:format","label":"Dc Format","values":["application/pdf"]},{"key":"dc:title","label":"Title","values":["Design of agentic ai systems for automated data extraction in medical processes /","Agentinių sistemų projektavimas automatizuotam duomenų išgavimui medicininiuose procesuose."]}]}],"canonical_facts":{"dc:creator":["Grenda, Marijonas,"],"dc:date":["2026"],"dc:description":["This thesis examines the design and evaluation of an agentic artificial intelligence system for automated data extraction from anonymized Lithuanian pathology reports. The goal is to convert unstructured clinical narrative into structured records that follow International Collaboration on Cancer Reporting (ICCR) datasets while preserving field-level traceability: every extracted value must be linked to a verbatim source span that supports it. To address this task, an agentic extraction system was developed in which an ICCR schema is populated through separate field-level processing steps. The system first retrieves source spans relevant to a field, then extracts a candidate value, verifies it using a Chain-of-Verification step, and emits schema-constrained JSON output. This architecture is compared against a strong one-shot baseline that receives the same case, schema, and field documentation. The evaluation uses a locked suite of 38 reviewed clinical cases covering four ICCR datasets: breast invasive carcinoma, radical prostatectomy, prostate biopsy, and colorectal resection. Both conditions use the same Qwen3 235B backbone model, and the evaluation is performed with the correct ICCR schema supplied by the suite manifest. The agentic system achieves 75.7% micro F1, compared with 63.0% for the one-shot baseline, and improves citation precision from 76.0% to 90.6%. The results show that the agentic control flow improves extraction mainly by reducing false-positive values and semantically unsupported source citations. The main remaining errors involve multi-value fields, numerical interpretation, and the scoring of clinically equivalent free-text paraphrases."],"dc:format":["application/pdf"],"dc:identifier":["https://repository.vu.lt/VU:ELABAETD308336925&prefLang=en_US"],"dc:language":["eng"],"dc:publisher":["Institutional Repository of Vilnius University"],"dc:relation":["https://epublications.vu.lt/object/elaba:308336925/308336925.pdf"],"dc:rights":["info:eu-repo/semantics/openAccess"],"dc:subject":["agentic AI systems, medical information extraction, pathology reports, ICCR, large language models, source traceability, citation precision"],"dc:title":["Design of agentic ai systems for automated data extraction in medical processes /","Agentinių sistemų projektavimas automatizuotam duomenų išgavimui medicininiuose procesuose."],"dc:type":["info:eu-repo/semantics/bachelorThesis"]},"updated_at":"2026-11-19T10:12:59Z"}