Optellum’s AI could speed up lung cancer screening by detecting benign perifissural lung nodules

Optellum has been collaborating with Oxford University Hospitals and UMC Groningen to evaluate the ability of our AI to assist in lung cancer screening workflows.

A paper published in European Radiology illustrates the results of the collaboration between Optellum, Oxford University Hospitals NHS Foundation Trust (OUH) and University Medical Centre Groningen (UMCG). The main data used for the current study were collected as part of the public-private grant provided by the European Institute of Innovation and Technology (EIT).

UMCG is one of the largest hospitals in the Netherlands and our collaborators are the principal Radiological investigator and other members of the team behind the NELSON study, the second largest screening trial in the world, whose successful result saw lung cancer screening implemented across Europe. OUH is a world renowned center of clinical excellence and one of the largest NHS teaching trusts in the UK, partnered with Optellum in projects like LUCINDA, funded by EIT Health and IDEAL, funded by the National Institute for Health Research (NIHR).

In recent years there has been a growing interest in the use of computed tomography (CT) scans to screen patients for lung cancer. This has resulted in increased detection of false-positive results after detecting small to intermediate sized lung nodules, most of which are benign. False positives are a problem in lung cancer screening, requiring healthy people to undergo unnecessary procedures such as biopsies, PTE and CT scans. Perifissural nodules (PFN) are a sub-group of small-to-intermediate-sized solid nodules, which account for approximately 20–30% of all solid pulmonary nodules found in the lung cancer screening setting as well as incidentally in the clinical setting.1 Although some PFNs have rapid growth, similar to malignant nodules, they are benign in all typical cases. Automated reliable identification of these nodules could reduce the workload for radiologists and prevent unnecessary followup scans, as well as unnecessary worry for the patient.

Data from two regional health systems in the UK and Netherlands were collected in a retrospective study, comprising 1668 unique pulmonary nodules from 1260 individuals. Nodules were classified into subtypes including typical PFNs, by radiologists on-site, and were reviewed centrally to ensure consistency. The dataset was divided into a training set (1472 nodules, 1103 individuals) and an independent test set (196 nodules, 158 individuals). The training set data was used to train a convolutional neural network (CNN) developed by Optellum to classify nodules as either PFN or not, and the independent test set to validate its performance. The Optellum CNN was compared against a panel of clinicians comprising two radiologists with 8 and 10 years experience and a 4th year radiology resident. They were asked to identify the nodules in the test set as PFN or not, and then these results were compared with the classification by Optellum’s algorithm. The artificial intelligence demonstrated excellent performance and an agreement with the consensus of three radiologically trained readers, comparable to the agreement between trained readers.

This could have an impact on patient care in that radiologist time could be saved and patients spared unnecessary investigation of nodules that are not malignant, enabling faster and more accurate screening.

The lead authors of the resulting publication in European Radiology are core members of the NELSON European lung screening trial team, including Dr. Marjolein Heuvelmans (pulmonologist) and Prof. Matthijs Oudkerk (radiologist and Co-Principal Investigator of NELSON).

Dr Marjolein Heuvelmans, AIOS longziekten at Medisch Spectrum Twente, commented:
“It is good to see these results which indicate that artificial intelligence could be used to speed up lung cancer screening, reducing the workload of the radiologist and the anxiety of the patient.”

Read the paper here: Evaluation of a novel deep learning–based classifier for perifissural nodules


1. de Hoop B, van Ginneken B, Gietema H, Prokop M (2012) Pulmonary perifissural nodules on CT scans: rapid growth is not a predictor of malignancy. Radiology 265:611–616. https://doi.org/10.1148/radiol.12112351,
Ahn MI, Gleeson TG, Chan IH et al (2010) Perifissural nodules seen at CT screening for lung cancer. Radiology 254:949–956. https://doi.org/10.1148/radiol.09090031,
Mets OM, Chung K, Scholten ET et al (2018) Incidental perifissural nodules on routine chest computed tomography: lung cancer or not? Eur Radiol 28:1095–1101. https://doi.org/10.1007/s00330-017-5055-x