A 3D Probabilistic Deep Learning System for Detection and Diagnosis of Lung Cancer Using Low-Dose CT Scans
Problem
Detecting lung nodules and diagnosing malignancy accurately from low-dose CT screening scans.
Why it matters
Lung cancer screening at scale affects patient outcomes through early detection and diagnosis.
What the AI did
An end-to-end system of 3D convolutional neural networks couples nodule detection with malignancy classification, eliminating a false-positive reduction stage, and characterizes model uncertainty to produce calibrated probabilities.
Results
State-of-the-art performance on LUNA16 and Kaggle Data Science Bowl for both nodule detection and malignancy classification; no numeric metrics in abstract.
Impact
Peer-reviewed system with state-of-the-art public benchmark performance; clinical adoption not stated in source.
Record
| Venue | IEEE Transactions on Medical Imaging |
|---|---|
| Published | 2019 |
| Work type | article |
| Authors | Onur Özdemir, Rebecca L. Russell, Andrew A. Berlin |
| Citations | 265 (OpenAlex, at import time) |
| References | 47 |
| Model or method | 3D convolutional neural networks |
| Organization | not stated in source |
| Primary source | https://doi.org/10.1109/tmi.2019.2947595 |
| Open access | hybrid |
| Keywords | lung cancer, low-dose CT, 3D CNN, nodule detection, malignancy classification, uncertainty |
Importance rating
major — millions of lives, an industry or a field
Lung cancer screening/diagnosis at scale via low-dose CT; early detection affects patient outcomes, but abstract gives no scale numbers.