LungNet: How AI Can Diagnose Lung Cancer from CT Scans and Wearable Sensors

By Nuruzzaman Faruqui

·

Lung cancer is one of the deadliest cancers, yet it is often curable when caught early. LungNet is an AI system that reads CT scans together with data from wearable health sensors, so it can spot lung cancer and even tell apart its earliest stages.

96.81%accuracy across five lung cancer classes
91.6%accuracy telling apart early sub-stages 1A to 2B
525,000images used to train the model

The problem

Doctors look for lung cancer in CT scans, which are detailed X-ray images of the chest. The trouble is that early tumours, called nodules, can look a lot like harmless spots. Even experienced radiologists find some of them hard to judge, and many parts of the world have too few specialists to read every scan quickly.

Computers can help, but a scan alone does not tell the whole story. A patient’s breathing, heart rate and other body signals also carry clues about how sick they are.

What the research did

We built LungNet, a deep learning model with a custom 22-layer convolutional neural network, the kind of AI that is very good at finding patterns in images. What makes LungNet different is that it learns from two kinds of information at once: CT scan images and physiological readings collected by wearable medical sensors, part of what is called the medical Internet of Things.

Think of it like a doctor who looks at your scan while also checking your vital signs, rather than judging from the picture alone. LungNet blends the patterns it finds in both and runs from a central server, so a clinic anywhere with an internet connection could send data to it.

What we found

Trained on a balanced set of 525,000 images, LungNet sorted lung cancer into five classes with 96.81% accuracy and a false positive rate of just 3.35%, which means it rarely raised a false alarm. It outperformed similar image-only AI classifiers.

It also went a step further than most systems: it separated stage 1 and stage 2 cancers into the finer sub-stages 1A, 1B, 2A and 2B with 91.6% accuracy. Sub-stages matter because they guide treatment decisions.

Who benefits

  • Patients, through earlier and more consistent detection, when treatment works best
  • Radiologists, who get a reliable second opinion on hard-to-read scans
  • Hospitals in low-resource regions, including in Bangladesh, where a central AI service can support clinics without specialists on site
  • Researchers building the next generation of diagnostic AI: with 217 citations, LungNet is my most-cited paper

Limits and what comes next

LungNet was tested on research datasets, not in a live hospital, and any medical AI needs clinical trials and regulatory approval before doctors can rely on it. It is designed to support specialists, not replace them. The next step is testing on data from more hospitals and patient groups to check that it works equally well for everyone.

The team

This research was led by me with Mohammad Abu Yousuf (Jahangirnagar University), Md Whaiduzzaman and Alistair Barros (Queensland University of Technology), AKM Azad and Mohammad Ali Moni, spanning Bangladesh and Australia.

Frequently asked questions

Can AI diagnose lung cancer?

AI can analyse CT scans and flag likely lung cancer with high accuracy. LungNet reached 96.81% in testing. But AI is a support tool: a qualified doctor still makes the diagnosis.

Why combine CT scans with wearable sensor data?

A scan shows what a tumour looks like, while body signals show how the patient is doing. Combining them gives the AI more evidence, much as a doctor considers both.

Is LungNet used in hospitals?

Not yet. It has been validated on research data. Clinical trials and regulatory approval would be needed before routine use.

About the author

Nuruzzaman Faruqui is an award-winning AI educator and researcher who has taught artificial intelligence and machine learning at university level since 2018. A two-time winner of the Excellence in Teaching Award (2023, 2024) and winner of the Best Researcher Award (2024) at Daffodil International University, he is now a doctoral researcher in AI at Monash University, Australia. He has published 53 papers and six books, and writes these articles to make AI research easy for everyone to understand. His teaching · About Zaman · Get in touch

Keep reading