New AI Tool for Brain Tumor Diagnostics
A team at the Research Centre Jülich from the Helmholtz Research Field Information has developed a deep learning algorithm that can automatically detect and assess brain tumors in PET scans. The AI achieves comparable results to an experienced doctor in a fraction of the time, the team reports in the Journal of Nuclear Medicine. (Source: Jülich Research Centre – Press Releases)
During the initial diagnosis and treatment, it’s crucial to determine the extent and volume of a brain tumor as accurately as possible. Can the tumor be operated on? Is it responding to treatment or is it still growing? The method of choice initially is Magnetic Resonance Imaging (MRI) because it accurately captures structural changes. However, structural anomalies do not necessarily reflect the actual size of the tumor. In many cases, a version of Positron Emission Tomography (PET) provides results different from MRI, as studies have shown. Evaluating PET scans concerning tumor extent is time-consuming, which is why it’s rarely done routinely in clinics. Philipp Lohmann’s team from the Institute for Neuroscience and Medicine (INM-4) at the Research Centre Jülich has now developed a new deep learning algorithm that automatically detects tumors in the PET data and determines their volume. This makes the analysis much faster without compromising the quality of the results. “The AI tool we developed could help physicians get important diagnostic information that is rarely available,” says Lohmann.
PET uses radioactively labeled biomolecules to visualize specific metabolic processes. Especially radioactively labeled amino acids have proven effective for displaying brain tumors. Rapidly growing cancer cells absorb the amino acids much faster than healthy brain cells. “On the PET images, you can often detect a different or larger tumor extent based on the enriched amino acids than with MRI,” Lohmann explains. Experts refer to this as the metabolic tumor volume since the measurement method is based on cell metabolic processes. However, a doctor has to identify the tumor contours in several dozen layers of a PET scan, which is time-consuming. “That’s why this method, despite its high informative value, is rarely used in clinical practice,” Lohmann knows.
To change this, the team developed a deep learning algorithm called “JuST BrainPET” (Juelich Segmentation Tool for Brain Tumor PET) that fully automatically evaluates amino acid PET images for the metabolic tumor volume. The algorithm was trained with 476 PET datasets and then tested on 223 that were all part of an initial diagnosis or follow-up examinations of a total of 555 brain tumor patients. The related information about the tumor volume came from the analyses of experienced nuclear medicine physicians.
“Our AI has learned to distinguish the tumor from other structures that also absorb amino acids for physiological reasons, such as vessels or muscle tissue,” says Robin Gutsche, a doctoral student at INM-4 and the lead author of the study, who played a significant role in the development of the AI algorithm. “This works even when the tumors are directly adjacent to these structures.” Within a few minutes, the algorithm detects the tumor and determines the metabolic tumor volume. The results are very consistent with the values determined by the experts.
The research team also verified the clinical utility of the algorithm by having it assess the treatment success of chemotherapy in patients with gliomas. “The goal was to answer questions such as: How well does a patient respond to therapy, or what is his prognosis?” says Lohmann. “We were able to show that the AI performs the clinical assessment just as well as a specialist – but in a fraction of the time.” However, the AI is not meant to replace the doctor but to support them, he emphasizes.
Furthermore, the new AI approach could help standardize the evaluation of PET measurements to make the results more comparable across different clinics and institutions. “We hope our algorithm encourages treating physicians in neuro-oncological centers to use the amino acid PET more frequently for their patients – even if they have little experience with the method,” says Lohmann. JuST_BrainPET is available for free on the Github platform.”
FZJ/B. Schunk, 24.10.2023
Note: The article has been translated from German to English. It is based on a press release from the Research Centre Jülich.
The original press release can be found at:
Neues KI-Werkzeug für die Hirntumordiagnostik (only in german)
Related Links:
JuST BrainPET: https://github.com/MIC-DKFZ/nnUNet/tree/nnunetv1#useful-resources
Localization in the Helmholtz Research Field Information:
Helmholtz Research Field Information, Program 2: Natural, Artificial and Cognitive Information Processing, Topic 5: Decoding Brain Organization and Dysfunction
Contact:
Priv.-Doz. Dr. Philipp Lohmann
Institute of Neuroscience and Medicine (INM)
Medical Imaging Physics (INM-4)
Forschungszentrum Jülich
Phone: +49 2461/61-96357
E-Mail: p.lohmann@fz-juelich.de
Contact for this Press release:
Dr. Barbara Schunk
Press officer
Forschungszentrum Jülich
Phone: +49 2461 61-8031
E-Mail: b.schunk@fz-juelich.de
