For years, AI in healthcare felt like a promise perpetually stuck in the research lab, but a new wave of open-source projects and peer-reviewed studies suggests the field is finally getting serious about cancer.
We’re no longer just seeing generic chatbots repurposed for medicine; instead, developers and researchers are building specialized, multi-agent systems that tackle the brutal complexity of oncology head-on – from automating GRADE-based literature reviews and parsing messy pathology reports to matching patients with clinical trials across 14 countries in real time.
This post rounds up eight of the most compelling tools, libraries, and papers emerging right now, including a Nature Cancer study showing agentic AI can hit 87% accuracy on complex treatment plans and an AWS architecture proving HIPAA-compliant multi-agent workflows are production-ready.
Crucially, every single one of these projects keeps a human in the loop, because while AI can compress years of uncertainty into seconds, the final call on a patient’s life still belongs to a doctor.
Here are the best open-source AI Cancer Projects
1- Oncology Clinical Decision Support System
Oncology treatment decisions are brutally slow – doctors juggle literature reviews, survival statistics, and evidence grading by hand, and there’s always a nagging fear of missing something published last week.
So I built a multi-agent AI system (running on Gemini 2.5 Flash) where specialized agents handle each piece in parallel: one digs through PubMed and Google Scholar, another crunches Kaplan-Meier curves and log-rank tests, and a third drafts GRADE-based recommendations with confidence scores.
The whole thing still funnels through a “Doctor-in-the-Loop” approval step, because no matter how fast the AI works, a real physician signs off before any recommendation touches a patient.
2- Oncoteam
When someone you love has cancer, you quickly realize that even the best oncologist only has fifteen minutes per visit and can’t possibly track every new trial or lab trend in real time.
Oncoteam is an open-source AI ally that lives on your phone and turns overwhelming medical data into clarity – translating labs, visualizing treatment trends, and prepping personalized questions before every appointment.
It actively searches clinical trial registries across the US and 14 European countries, then filters them through an AI funnel so you actually know which ones are relevant instead of drowning in noise.
You can also get a literature-backed second opinion cross-referenced against NCCN guidelines and DeVita’s textbooks, not to undermine your doctor but to make sure no option slips through the cracks.
Everything is accessible via WhatsApp and includes compassionate family summaries, because advocacy shouldn’t require a medical degree or wait until the next scheduled visit.
3- Onco-Logic: An Integrated AI Suite for Precision Oncology
Onco-Logic is an open-source AI suite that brings together three critical pieces of cancer data, pathology reports, gene expression, and survival outcomes, into one unified platform for precision oncology.
Its NLP engine reads messy free-text pathology reports and turns them into structured, computable data while generating plain-language prognostic summaries clinicians can actually use at the bedside.
The genetics module tackles the “high-dimension, low-sample-size” problem head-on, using machine learning and SHAP explanations to automatically subtype tumors and highlight the specific genes driving each classification.
Finally, its breast cancer survival predictor handles severe class imbalance with advanced resampling to deliver 95.9% accuracy, giving doctors an interactive risk tool backed by transparent, explainable models rather than a black box.
4- OncoAgent: Multi-Agent Oncology Triage System – AMD Hackathon 2026
A developer built OncoAgent after watching his mother endure two years of misdiagnosis, driven by the reality that in oncology, delays like that can be the difference between life and death.
It’s an open-source, multi-agent AI system optimized for AMD MI300X hardware that uses a strict LangGraph workflow with dedicated router, specialist, and critic agents to keep every output grounded in official NCCN and ESMO guidelines.
The system actively refuses to guess when evidence is missing, leveraging adaptive semantic chunking and a dual-tier Qwen model setup to deliver fast, hallucination-resistant triage support for clinicians.
He just launched a live Gradio demo on Hugging Face Spaces for the AMD Developer Hackathon and is asking the community to drop a like to help get this life-saving tool in front of more people.
5- Oncology AI Fact Sheet
Researchers created a simple, fillable fact sheet template to help oncology clinicians quickly understand how to safely use AI tools in their daily practice.
The sheet focuses on practical, non-technical details like what the AI does, who should use it, and whether it has FDA approval or real-world validation.
It clearly explains what data the system needs as input, what output to expect, and how confident the algorithm is in its predictions.
Users also learn if the AI provides explanations for its decisions and exactly how much human oversight is required before acting on results.
A completed example for a lung tumor segmentation tool shows how each field should be filled out to keep information consistent and easy to digest at the point of care.
6- OpenClaw Medical: 869-Skill Library for Medical AI Agents
OpenClaw Medical is a massive open-source library of 869 AI skills that includes dedicated oncology tools for matching targeted cancer therapies, parsing tumor mutations, and analyzing single-cell tumor biology.
However, because these tools handle life-or-death clinical decisions, developers are strictly warned to build their own safety guardrails and patient privacy filters before deploying them in real cancer care.
7- Development and validation of an autonomous artificial intelligence agent for clinical decision-making in oncology
A new Nature Cancer research paper shows that giving GPT-4 a toolkit of specialized oncology apps, like image segmenters, mutation predictors, and guideline databases, turns it from a generic chatbot into an autonomous clinical agent that gets treatment plans right 87% of the time.
While plain GPT-4 only nailed 30% of complex multimodal cancer cases, this agentic system autonomously chains together radiology scans, histopathology slides, and live literature searches to reason through patient data just like a human specialist would.
The catch is that open-weight models like Llama-3 and Mixtral completely crashed when asked to use these medical tools, proving that for now, building safe AI oncologists still depends heavily on proprietary models with rock-solid function-calling skills.
8- Accelerating cancer treatment planning with agentic AI with AWS
This AWS technical guide shows how to build a multi-agent AI system that accelerates breast cancer treatment planning by having specialized bots work in parallel instead of waiting on rigid, step-by-step workflows.
A central orchestrator delegates tasks to three dedicated agents, one pulls multimodal patient data from HealthLake and HealthImaging, another extracts relevant clinical history, and a third fetches the latest ASCO/NCCN guidelines and clinical trials.
Everything runs on Amazon Bedrock’s new AgentCore Runtime with strict HIPAA-compliant authentication, ensuring that sensitive patient data stays secure while the AI gathers real-time context. The oncologist always has the final say through a human-in-the-loop review, but the heavy lifting of synthesizing imaging, records, and research into a draft treatment plan happens in seconds rather than hours.
9- Onco-Sentinel AI
Onco-Sentinel AI is a clinical intelligence platform built to tackle the growing complexity of cancer care by unifying medication safety, toxicity surveillance, and pharmacokinetic data into one cohesive system.
Inspired by real-world challenges in adult and pediatric oncology, it integrates patient history, genomic information, and drug exposure analytics to support safer, more personalized treatment decisions across the entire care continuum.
The platform aims to bridge fragmented clinical systems by connecting real-world evidence with precision oncology research, ultimately helping multidisciplinary teams monitor patients longitudinally and prevent adverse drug events.
10- EvoMDT: a self-evolving multi-agent system for structured clinical decision-making in multi-cancer
EvoMDT is a self-evolving multi-agent system designed to replicate the structured reasoning of multidisciplinary tumor boards, using expert feedback and outcome signals to continuously refine its prompts and consensus protocols without losing traceability.
Unlike static LLMs that produce opaque answers, it performs domain-specific inference at the lesion level with enforced safety gates and auditable conflict resolution, achieving decision quality comparable to human MDTs while cutting response time by 30–40%.
Evaluated across six public benchmarks and four real-world cancer datasets, EvoMDT outperformed frontier models like Claude-3 and Med-PaLM 2 in guideline concordance and semantic alignment, proving that adaptive AI can scale trustworthy, evidence-linked precision oncology without sacrificing clinical rigor.
11- Awesome AI Cancer
This open-access repository serves as a comprehensive, curated index of AI and deep learning resources for oncology, covering everything from code and datasets to peer-reviewed papers across brain, breast, lung, prostate, and pancreatic cancers.
It organizes materials by clinical application, including detection, prognosis, drug design, radiation therapy, and surgery, while linking directly to major benchmarks like CAMELYON17, TCGA, and nnU-Net so researchers can reproduce or build upon validated work.
Licensed under MIT/CC-BY for unrestricted reuse, the list functions as both a launchpad for new projects and a living archive of state-of-the-art methods, ensuring that advances in computational oncology remain transparent, reproducible, and freely accessible to the global research community.
