In the field of oncology, evaluating the effectiveness of chemotherapy in patients with bone cancer is a fundamental determinant of prognosis. A research team at Johns Hopkins Medicine has recently pioneered a groundbreaking advance in this field. They have successfully developed and trained a machine learning model to calculate percent necrosis (PN), a crucial metric that indicates the extent of necrosis. tumor death in patients with osteosarcoma. This innovative model demonstrates an impressive 85% accuracy compared to results obtained by a musculoskeletal pathologist. By removing a single outlier, the accuracy rises to an astonishing 99%.
Traditionally, the process of calculating NP has been laborious and dependent on extensive annotation data from musculoskeletal pathologists. Furthermore, it suffers from low interobserver reliability, so two pathologists analyzing the same whole-slide images (WSI) may reach different conclusions. Recognizing these challenges, the researchers highlighted the need for an alternative approach.
The team’s search led them to develop a weakly supervised Machine learning model that requires minimal annotation data for training. This innovative methodology means that a musculoskeletal pathologist using the model for NP calculation would only need to provide partially annotated WSIs, substantially reducing the pathologist’s workload.
To build this model, the team selected a comprehensive data set, including WSI, from the pathology archives of the distinguished Johns Hopkins tertiary cancer center in the US. These data exclusively comprise cases of intramedullary osteosarcoma, originating in the core of the bone , in patients who underwent both chemotherapy and surgery at the center between 2011 and 2021.
A musculoskeletal pathologist meticulously scored three distinct tissue types in each WSI collected: active tumor, necrotic tumor, and nontumor tissue. Additionally, the pathologist estimated the PN of each patient. Armed with this valuable information, the team embarked on the training phase.
The researchers explained the training process. They decided to train the model by teaching it to recognize image patterns. The WSIs were segregated into thousands of small patches and then divided into groups based on how the pathologist labeled them. Finally, these grouped patches were incorporated into the training model. This approach was chosen to provide the model with a more robust framework, avoiding the potential oversight that could occur by feeding it only a large WSI.
After training, the model and the musculoskeletal pathologist were presented with six WSIs to evaluate two patients with osteosarcoma. The results were notable, with an 85% positive correlation between the model’s NP calculations and tissue labeling compared to the pathologist’s findings. The only caveat arose from occasional difficulties in adequately identifying cartilage tissue, which led to an outlier due to the abundance of cartilage in a WSI. After its removal, the correlation shot up to an impressive 99%.
Going forward, the team plans to incorporate cartilage tissue into the model training and expand the scope of WSI to encompass various types of osteosarcoma beyond intramedullary. This study represents a significant step toward revolutionizing the evaluation of osteosarcoma treatment outcomes.
Review the Paper. All credit for this research goes to the researchers of this project. Also, don’t forget to join. our 32k+ ML SubReddit, 41k+ Facebook community, Discord channel, and Electronic newsletterwhere we share the latest news on ai research, interesting ai projects and more.
If you like our work, you’ll love our newsletter.
we are also in Telegram and WhatsApp.
Niharika is a Technical Consulting Intern at Marktechpost. She is a third-year student currently pursuing her B.tech degree at the Indian Institute of technology (IIT), Kharagpur. She is a very enthusiastic person with a keen interest in machine learning, data science and artificial intelligence and an avid reader of the latest developments in these fields.
<!– ai CONTENT END 2 –>
