Newswise — Winner of the 2018 Ronald L. Bittner Award on Brain Tumor Research, Todd Hollon, MD, presented his research, Convolutional Neural Networks Provide Rapid Intraoperative Diagnosis of Neurosurgical Specimens Imaged with Stimulated Raman Histology, during the 2018 American Association of Neurological Surgeons (AANS) Annual Scientific Meeting.
Intraoperative diagnosis is essential in the surgical management of brain tumors. Stimulated Raman histology (SRH) uses the intrinsic biochemical properties of fresh, unprocessed surgical specimens to provide label-free digital histologic images, eliminating the need for a conventional histology lab for intraoperative diagnosis. SRH is an ideal imaging modality to implement machine-learning strategies for tissue diagnosis due to the robust histochemical information encoded in pixel data.
For this project, researchers developed a deep convolutional neural network (CNN), which provides rapid, standardized diagnosis of fresh brain specimens. For CNN training, 20,000 400x400μm SRH fields of view (FOV) from 270 patients across seven common intraoperative diagnostic categories were used. The neural network was developed from the GoogleNet InceptionV3 CNN architecture, which includes 24 million trainable parameters. Model validation was completed on 3,900 FOVs from 70 patients. The model was evaluated on its ability to distinguish 1) diagnostic versus non-diagnostic tissue, 2) tumor versus non-tumor tissue, 3) glial versus non-glial tumors and 4) provide the correct intraoperative diagnostic category.
The trained CNN differentiated 1) diagnostic from non-diagnostic FOVs with 99.1 percent accuracy, 2) tumor from non-tumor FOVs with 98.4 percent accuracy and 3) glial from non-glial tumor FOVs with 96.4 percent accuracy. Using CNN classifier probabilities, the area under the curve on ROC analysis for the above listed metrics was 0.992, 0.985 and 0.985, respectively. By compiling FOV data from each specimen, the CNN predicted the correct intraoperative diagnostic category in 98.6 percent (69/70) of cases in the validation set. The only classification error by the CNN was an instance of dense gliotic brain tissue incorrectly labeled low-grade glioma.
This study demonstrates the feasibility of applying deep machine learning for intraoperative diagnosis of neurosurgical specimens. SRH and convolutional neural networks may ultimately be used to guide decision-making during brain tumor surgery independent of conventional neuropathology resources.
Author Block: Daniel Orringer, MD, FAANS
Disclosure: Dr. Hollon has no disclosures. Dr. Orringer is an advisor and shareholder in Invenio Imaging Inc., a company devoted to the commercialization of stimulated Raman histology.
Media Representatives: The 2018 AANS Annual Scientific Meeting website’s press section will include releases on highlighted scientific research, AANS officers and award winners, Neurosurgery Awareness Month and other relevant information about the 2018 program. Releases will be posted on the 2018 AANS Annual Scientific Meeting website. If you have interest in a topic related to neurosurgery or would like to interview a neurosurgeon — either onsite or via telephone — during the event, please contact Alice Kelsey, AANS associate executive director, via email at [email protected].
About the 2018 AANS Annual Scientific Meeting: Attended by neurosurgeons, neurosurgical residents, medical students, neuroscience nurses, clinical specialists, physician assistants, allied health professionals and other medical professionals, the AANS Annual Scientific Meeting is the largest gathering of neurosurgeons in the nation, with an emphasis on the field’s latest research and technological advances. The scientific presentations accepted for the 2018 event will represent cutting-edge examples of the incredible developments taking place within the field of neurosurgery. Find additional information about the 2018 AANS Annual Scientific Meeting and the meeting program here.
Founded in 1931 as the Harvey Cushing Society, the American Association of Neurological Surgeons (AANS) is a scientific and educational association with more than 11,000 members worldwide. The AANS is dedicated to advancing the specialty of neurological surgery in order to provide the highest quality of neurosurgical care to the public. Fellows of the AANS are board-certified by the American Board of Neurological Surgery, the Royal College of Physicians and Surgeons of Canada or the Mexican Council of Neurological Surgery, A.C. Neurosurgery is the medical specialty concerned with the prevention, diagnosis, treatment and rehabilitation of disorders that affect the spinal column, spinal cord, brain, nervous system and peripheral nerves.
For more information, visit www.AANS.org.
RSS