Researchers have developed and trained a new machine learning model to calculate percent necrosis in patients with osteosarcoma, according to a novel study published by LiBrizzi et al in the Journal of Orthopaedic Research.
Background
A postchemotherapy percent necrosis calculation often helps provide patients with a prognosis for survival. For instance, a percent necrosis of 99% indicates that the tumor is dead, the chemotherapy was effective, and the patient has improved odds of surviving. Pathologists can calculate percent necrosis by interpreting and annotating whole-slide images—thinly sliced sections of a bone tissue specimen that are mounted onto slides for microscopic analysis.
“Calculating the [percent necrosis] is a labor-intensive process that requires a lot of annotation data from the musculoskeletal pathologist,” explained co–lead study author Christa LiBrizzi, MD, a resident in the Department of Orthopaedic Surgery at Johns Hopkins Medicine. “Additionally, it has low interobserver reliability, meaning that two pathologists trying to calculate a [percent necrosis] from the same [whole-slide images] will often report different conclusions. Due to these factors, we thought trying to calculate a [percent necrosis] by alternate means would be a worthwhile effort,” she added.
Study Methods and Results
In the new study, the researchers sought to develop a weakly supervised machine learning model that required minimal annotation data for training. They noted that training the model in this way would allow a pathologist using the novel technology to calculate a patient’s percent necrosis to provide it only with partially annotated whole-slide images—thus reducing the pathologist’s labor burden.
Initially, the researchers used the pathology archives at the Johns Hopkins’ U.S. tertiary cancer center to gather data, including whole-slide images, from patients with intramedullary osteosarcoma who had undergone chemotherapy and surgery from 2011 and 2021. They then asked a musculoskeletal pathologist to partially annotate three types of tissue on each of the whole-slide images: active tumor, dead tumor, and nontumor tissue. The pathologist also estimated the percent necrosis for each patient. Using this information, the researchers were able to train the model.
“We decided to train the model by teaching it to recognize image patterns,” revealed co–lead study author Zhenzhen Wang, a doctoral student in the Department of Biomedical Engineering at the Johns Hopkins University School of Medicine. “We segregated the [whole-slide images] into thousands of small patches, then divided the patches into groups based on how they were labeled by the pathologist. Finally, we fed these grouped patches into the model to train it. We thought this would give the model a more robust frame of reference than simply feeding it one large [whole-slide image] and risking missing the forest for the trees,” she continued.
After training the machine learning model, the researchers assigned the model and the musculoskeletal pathologist six whole-slide images to interpret from two patients with osteosarcoma.
The researchers found an 85% positive correlation between the model and the pathologist’s percent necrosis calculations and tissue labeling. The model did not always properly label cartilage, which led to an outlier as a result of an abundance of cartilage on one of the whole-slide images. However, upon removing the outlier, the researchers reported that the correlation increased to 99%.
Conclusions
“If this model were to be validated and produced, it could help expedite the evaluation of chemotherapy’s effectiveness on a patient—and thus, get them a prognosis estimate sooner,” emphasized Dr. LiBrizzi. “That would reduce health-care costs, as well as labor burdens on musculoskeletal pathologists,” she concluded.
In future studies, the researchers plan to include cartilage tissue in the model’s training and to diversify the whole-slide images to include other types of osteosarcoma beyond intramedullary.
Disclosure: For full disclosures of the study authors, visit onlinelibrary.wiley.com.
