A systematic review of autonomous multiagent frameworks for early breast cancer detection using thermography
This systematic review of 100 studies on AI-driven breast cancer thermography reveals that while deep learning models achieve high accuracy, the field is hindered by limited dataset diversity, a lack of clinical validation, insufficient explainability, and a critical absence of agentic or multi-agent AI frameworks.
Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
Imagine you are a detective trying to solve a mystery inside a city. Usually, detectives use flashlights and X-ray machines to look for clues, but those tools can be expensive, require special training, or even give off a tiny bit of radiation that isn't great for frequent use. Enter Thermography: think of it as a "heat camera." Instead of seeing light, it sees heat. Since cancerous cells often burn hotter and have more blood flow than healthy tissue, they show up as bright, warm spots on this heat map. It's like using a thermal night-vision goggles to spot a warm cookie in a cold kitchen.
For a long time, these heat cameras were a bit hit-or-miss. But recently, scientists started teaching computers to "see" these heat maps using Artificial Intelligence (AI). Specifically, they use Deep Learning, which is like a digital brain that learns by looking at thousands of pictures until it gets really good at spotting patterns. The big question researchers are asking is: Can we build a super-smart, automatic system that uses these heat cameras to find breast cancer early, safely, and cheaply, especially for places that don't have fancy hospitals?
This paper is a massive "report card" for 100 different studies that tried to answer that question between 2020 and 2026. The authors, Pradnya Narkhede and Davierwala Shehrevar, didn't just look at the numbers; they acted like a team of editors checking the homework of 100 different students. They wanted to see how good these AI systems really are, what tools they used, and—most importantly—what they are missing.
Here is the big story they found:
The "Super-Student" Trap
The paper discovered that many AI models are acting like "super-students" who have memorized the answers to a single, tiny practice test. In the world of thermography, there is one specific dataset called DMR-IR that contains images from only 287 patients. Shockingly, 73% of all the studies reviewed relied almost entirely on this single group of people. Because the AI has seen these specific images so many times, it gets incredibly high scores, claiming 95% to 99.9% accuracy.
However, the authors warn that this is like a student who memorized the answers to a math quiz but has never seen a real-world problem. The paper argues that these high scores are likely an illusion caused by the small, simple dataset, not because the AI is ready to walk into a real hospital and diagnose a real patient. When they looked for studies that tested the AI on new, diverse groups of people or in real-world clinics, the numbers dropped, and the evidence became much weaker.
The Missing "Team of Detectives"
The most exciting and surprising finding of this paper is a massive gap in the research. The authors looked for a specific type of advanced AI called Agentic AI or Multi-Agent Systems.
To understand this, imagine a regular AI as a single, very smart detective who looks at a photo and says, "This looks like cancer." Now, imagine an Agentic AI as a whole team of detectives working together. One detective checks the photo quality, another measures the heat patterns, a third checks the patient's medical history, a fourth explains why they think it's cancer, and a fifth writes the final report. These agents talk to each other, double-check their work, and make decisions together.
The paper found that while other types of AI (like those used for mammograms or general cancer research) have started using this "team of detectives" approach, zero studies have successfully applied this team-based, autonomous AI to breast thermography yet. It is a completely empty field. The authors suggest that the future of this technology isn't just about making a single model smarter, but about building this collaborative team (which they call the ATBIS framework) to handle the whole process from taking the picture to writing the report.
The Reality Check
The authors also pointed out some serious missing pieces in the current research:
- The "Why" is missing: About 60% of the studies didn't include any way to explain how the AI made its decision. In medicine, you can't just say "the computer said yes"; you need to know why so doctors can trust it.
- Real-world testing is rare: Out of 100 studies, only four were actual real-world clinical trials where patients were scanned in a hospital setting. Most were just computer simulations using old data.
- Privacy concerns: While some researchers are working on ways to train AI without sharing private patient data (called Federated Learning), none of these privacy-focused methods have been tested specifically on breast heat maps yet.
The Bottom Line
This paper concludes that while the technology to see heat patterns is getting very good at solving puzzles in a controlled lab, it hasn't quite learned how to solve the messy, real-world mystery of breast cancer yet. The high accuracy numbers are mostly a result of using the same small dataset over and over. The future, according to these researchers, lies in building that "team of AI detectives" (Agentic AI), gathering much larger and more diverse groups of patients to learn from, and finally testing these systems in real hospitals to see if they can actually save lives. Until then, the "super-student" is still just practicing for the big game.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.