OncoReasoning-3B-1225: Specialized AI for Cancer Clinical Trial Matching

OncoReasoning-3B-1225 is a 3.2 billion parameter model, distilled from gpt-oss-120b, and fine-tuned from a Llama 3.2-3B-Instruct base. It is specifically engineered to assist with the intricate process of cancer clinical trial matching, leveraging a maximum sequence length of 13000 tokens for comprehensive document analysis.

Key Capabilities

• Trial Space Extraction: Identifies target patient populations from eligibility criteria, including age, sex, cancer type, histology, prior treatments, and biomarkers.
• Patient Document Tagging: Processes pathology, imaging, and oncologist notes to extract and tag relevant clinical information (e.g., stage_at_diagnosis, biomarker, adverse_event) into JSON format.
• Patient History Summarization: Condenses concatenated patient clinical texts into structured summaries, focusing on cancer history, treatment, and potential boilerplate exclusion criteria.
• Trial Fit Assessment: Evaluates the suitability of a patient summary against defined trial spaces.
• Exclusion Criteria Screening: Screens patient summaries against common boilerplate exclusion criteria (e.g., uncontrolled brain metastases, heart failure, renal dysfunction).

Good For

This model is ideal for developers and researchers building applications that require precise, automated analysis of oncology clinical data for clinical trial recruitment, research, and decision support. Its specialized training makes it highly effective for tasks involving complex medical text and structured data extraction in the cancer domain. It is a research tool and not intended for standalone diagnostic use or standard clinical practice.