A new study has demonstrated how a novel diagnostic approach that combines advanced optical imaging with an artificial intelligence (AI) algorithm can generate accurate, real-time intraoperative diagnosis of brain tumors. The prospective study compared the diagnostic accuracy of stimulated Raman histology (SRH) brain tumor image classification through machine learning, with that of pathologist interpretation of conventional histologic images. The results indicated that diagnosis using the AI-based method was 94.6% accurate, with pathologist-based interpretation demonstrating 93.9% accuracy. The new system’s precise diagnostic capacity could be beneficial to centers that lack access to expert neuropathologists.
“As surgeons, we’re limited to acting on what we can see; this technology allows us to see what would otherwise be invisible, to improve speed and accuracy in the OR, and reduce the risk of misdiagnosis,” said senior author Daniel A. Orringer, MD, associate professor of neurosurgery at NYU Grossman School of Medicine, who helped develop SRH and who co-led the study with colleagues at the University of Michigan. “With this imaging technology, cancer operations are safer and more effective than ever before.”
Development of the AI algorithm, and the study in brain cancer patients, are described in Nature Medicine, in a paper titled, “Near real-time intraoperative brain tumor diagnosis using stimulated Raman histology and deep neural networks.”
About 80% of the 15.2 million people worldwide who are diagnosed with cancer every year will undergo surgery, and in many cases part of the removed tumor will be analyzed during surgery, partly to help provide a preliminary diagnosis, the authors explained. In the United States alone more than 1.1 million biopsy specimens are taken every year, and these must be interpreted by a contracting pathology workforce. However, the authors further commented, the existing workflow for intraoperative diagnosis based on hematoxylin and eosin staining of processed tissue is time, resource, and labor-intensive. “The conventional workflow for intraoperative histology, dating back over a century, necessitates tissue transport to a laboratory, specimen processing, slide preparation by highly trained technicians, and interpretation by a pathologist, with each step representing a potential barrier to delivering timely and effective surgical care,” they wrote.
The new approach developed by Orringer and colleagues exploits advances in optics and AI. The SRH imaging technique offers label-free, sub-micrometer-resolution images of unprocessed biologic tissues to reveal tumor infiltration. “SRH utilizes the intrinsic vibrational properties of lipids, proteins, and nucleic acids to generate image contrast, revealing diagnostic microscopic features and histologic findings poorly visualized with hematoxylin and eosin (H&E)-stained images, such as axons and lipid droplets, while eliminating the artifacts inherent in frozen or smear tissue preparations,” the investigators noted. The microscopic images are then processed and analyzed using an AI algorithm. Within two and a half minutes, surgeons can have a predicted brain tumor diagnosis. Using the same technology, after the resection, they are then able to accurately detect and remove otherwise undetectable tumor.
To build the artificial intelligence tool used for the study, the researchers trained a deep convolutional neural network (CNN) on more than 2.5 million samples from 415 patients to classify tissue into 13 histologic categories that represent the most common brain tumors, including malignant glioma, lymphoma, metastatic tumors, and meningioma. The CNN was then validated in a prospective clinical trial involving 278 patients undergoing brain tumor resection, or epilepsy surgery, at three university medical centers. Brain tumor specimens were biopsied from patients, split intraoperatively into sister specimens, and randomly assigned to the control or experimental arm. Specimens routed through the control arm of current standard practice, were transported to a pathology laboratory where they underwent specimen processing, slide preparation by technicians, and interpretation by pathologists. This overall process takes 20–30 minutes. Analysis of tissue samples in experimental arm was performed intraoperatively, from image acquisition and processing to diagnostic prediction via CNN. “Notably, the CNN was designed to predict diagnosis independent of clinical or radiographic findings, which were reviewed by study pathologists and are often of central importance in diagnosis,” the scientists wrote.
The results showed that the new technology was non-inferior to the current standard practice. “Overall diagnostic accuracy was 93.9% (261/278) for the conventional H&E histology arm and 94.6% (264/278) for the SRH plus CNN arm, exceeding our primary endpoint threshold for noninferiority (>91%).” Also of note, the data showed that diagnostic errors in the experimental group were distinct from errors in the control group, suggesting that a pathologist using the novel technique could achieve close to 100% accuracy.
“In conclusion, we have demonstrated how combining SRH with deep learning can be employed to rapidly predict intraoperative brain tumor diagnosis,” the authors commented. “Our workflow provides a transparent means of delivering expert-level intraoperative diagnosis where neuropathology resources are scarce, and improving diagnostic accuracy in resource-rich centers. The workflow also allows surgeons to access histologic data in near real-time, enabling more seamless use of histology to inform surgical decision-making based on microscopic tissue features.”
“SRH will revolutionize the field of neuropathology by improving decision-making during surgery and providing expert-level assessment in the hospitals where trained neuropathologists are not available,” noted study co-author Matija Snuderl, MD, associate professor in the department of pathology at NYU Grossman School of Medicine.
The study authors anticipate the development of AI algorithms that can predict key molecular changes in brain tumors. “Importantly, our AI-based workflow provides unparalleled access to microscopic tissue diagnosis at the bedside during surgery, facilitating detection of residual tumor, reducing the risk of removing histologically normal tissue adjacent to a lesion, enabling the study of regional histologic and molecular heterogeneity, and minimizing the chance of nondiagnostic biopsy or misdiagnosis due to sampling error.”
Implementation of the new system is the most recent of NYU Langone’s efforts to integrate artificial intelligence in clinical practice to improve diagnostics of cancer. Researchers and clinicians at NYU Langone’s Perlmutter Cancer Center have made recent strides in lung cancer, breast cancer, and brain tumors.
