Researchers at Weill Cornell Medicine have developed an innovative solution that can better predict the effect of chemotherapy in patients with muscle-invasive - where tumor tissue is also present in the muscle layer - bladder cancer. AI and machine learning technologies were used in the development of the new prediction model.
Existing models for determining treatment for muscle-invasive bladder cancer use a single data source to predict the outcome. By employing AI and machine learning technology, this new study was able to use all image data from the entire tumor and gene expression analyses. This resulted in the discovery of important genes and tumor characteristics that may determine the success of treatment.
Personalized treatment
The ability to accurately predict how a patient will respond to standard treatment for this type of cancer can help physicians personalize treatment. This may also potentially prevent patients in whom it appears that treatment will be effective from having to undergo bladder removal. “This work represents the spirit of precision medicine. We want to be able to apply the right treatment for the right patient at the right time,” said the researchers, whose work was recently published in Digital Medicine.
The Weill Cornell Medicine researchers are not the only ones harnessing the potential of AI and machine learning in the search for better cancer treatment forms and prediction models. A few weeks ago, we reported on German researchers who developed an AI model that improves the personalization of cancer treatments by analyzing 350 parameters, including medical history, laboratory values, imaging and genetic analyses.
Analyzing image data
To analyze the images, made available by the SWOG Cancer Research Network, the Weill Cornell Medicine researchers used specialized AI methods called graphical neural networks. These capture how cancer cells, immune cells and fibroblasts are organized and interact within the tumor. In addition, the researchers also deployed automated image analysis to identify these different cell types at the tumor site.
Combining the image-based input with the gene expression data to train and test their AI-driven deep learning model resulted in better predictions of clinical response than models that used only gene expression or imaging. “On a scale of 0 to 1, where 1 is perfect and 0 means nothing is correct, our multimodal model comes close to 0.8, while unimodal models that rely on only one data source get about 0.6. That in itself is promising, but we plan to hone the model for further improvements,” says Dr. Fei Wang, professor of health sciences at Weill Cornell Medicine.
The researchers plan to introduce more types of data into the model, such as mutational analyses of tumor DNA that can be picked up in blood or urine, or spatial analyses that can more accurately determine exactly what types of cells are present in the bladder.
