Project 7: Artificial Intelligence in Lung Cancer Detection and Staging using Ultralow-dose Positron Emission Tomography

The National Lung Cancer Screening Trial (NLST) reported that low-dose computed tomography (CT) based screening may decrease lung cancer-specific mortality. However, a CT-based screening program still results in a high frequency of false-positive findings with potentially unnecessary invasive diagnostic procedures. 18Ffluordeoxyglucose (FDG) positron emission tomography (PET) has been shown to increase the specificity by reliably detecting and characterizing pulmonary malignancy in vivo. One very important criterion for characterizing a pulmonary nodule as malignant (and justifying further invasive testing) is its FDG uptake. Therefore, FDG-PET/CT is recommended for further characterizing lung nodules detected incidentally on CT. Recent technical innovations allow for a significant dose reduction for clinical FDG-PET. The interest in computer-aided diagnosis for detection and classification of different oncologic diseases has grown in recent years, and there is an increasing interest in applying deep learning (e.g. deep convolutional neural networks) in nuclear medicine and radiology. AI techniques my improve on diagnosis and staging of lung cancer using ultralow-dose FDG-PET datasets which are acquired using a novel digital PET/CT scanner equipped with silicon-based photomultiplier technology and time-of-flight capability.

1. Using AI techniques, FDG-avid and therefore potentially malignant pulmonary nodules may be classified in a standardized and observer-independent automated and robust way
2. Using AI techniques, lung cancer staging in a standardized and observer-independent way using FDG-PET will improve the accuracy of the diagnosis.
3. A standardized assessment of lung cancer using FDG-PET/CT will eventually result in a lower number of false-positives and false-negatives (e.g., prevent unnecessary biopsies and overlooked malignant lesions), thereby lowering healthcare costs and improving patient care.

The aim is to develop an AI software tool capable to

1. automatically classify pulmonary nodules based on their FDG-PET characteristics,
2. automatically detect additional lesions and stage lung cancer patients,
3. write a standardized report.