AI Tool for PET Imaging Enables Fully Automated Detection and Evaluation of Brain Tumors

Positron Emission Tomography (PET) is becoming a critical tool for diagnosing brain tumors, adding to the insights given by traditional MRI scans. In recent years, numerous studies have revealed the utility of evaluating metabolic tumor volume to gauge the effectiveness of treatments for brain tumors. However, such evaluations usually take a lot of time and hence are not commonly performed in regular clinical settings. Now, a new artificial intelligence (AI) tool offers an automated, simple, and objective method to identify and assess brain tumors. Designed to work with amino acid PET scans, this deep-learning algorithm can also quickly evaluate a patient’s response to treatment with the same level of accuracy as a seasoned doctor.

This deep-learning-based segmentation algorithm for the comprehensive and automated volumetric assessment of amino acid PET scans has been developed by a team of researchers the Institute of Neuroscience and Medicine (INM, Juelich, Germany). The researchers have also tested its efficacy for evaluating treatment responses in patients with gliomas. The team analyzed 699 18F-FET PET scans (either initial or follow-up) taken from 555 individuals with brain tumors. The algorithm was configured using both training and test datasets, and the changes in metabolic tumor volumes were measured.

Moreover, the algorithm was applied to data from a recently released 18F-FET PET study that examined the treatment responses in glioblastoma patients who underwent adjuvant temozolomide chemotherapy. The algorithm's evaluation was then compared to the judgment of a skilled physician, as documented in that study. Within the test dataset, the algorithm accurately identified 92% of the lesions that showed increased uptake and 85% of the lesions with isometric or hypometabolic uptake. The algorithm-detected changes in metabolic tumor volume significantly aligned with predictions of disease-free and overall survival rates, confirming the observations made by the physician. To aid its adoption in clinical settings, this segmentation algorithm is openly accessible and can be run on a standard GPU-equipped computer in less than two minutes without requiring any preprocessing.

“These findings highlight the value of the deep learning-based segmentation algorithm for improvement and automatization of clinical decision-making based on the volumetric evaluation of amino acid PET,” said Philipp Lohmann, PhD, assistant professor (Habilitation) in Medical Physics, and team leader for Quantitative Image Analysis & AI at the INM. “The segmentation tool developed in our study could be an important platform to further promote amino acid PET and to strengthen its clinical value, which may give brain tumor patients access to important diagnostic information that was previously unavailable or difficult to obtain.”

Related Links:
INM