The AI ​​model powered by AI improves the anticipation

Using the power of artificial intelligence and machine learning technology, scientists from Weill Cornell Medicine have developed a more effective prediction model how patients with bladder bladder cancer will react to chemotherapy. The model uses the data of the whole tumor imaging and the analysis of gene expression in a way that exceeds previous models using one type of data.

The study, published on March 22, identifies key genes and tumor characteristics that can determine the success of treatment. The ability to accurately predict how the individual will react to standard therapy of the care of this malignant cancer can help doctors personalize treatment and potentially save those who react well from the removal of the bladder.

“This work represents the spirit of precise medicine,” said Dr. Fei Wang, professor of population health sciences at Weill Cornell Medicine and director, the founder of the Institute of Artificial Intelligence for Digital Health, which conducts the study.

“We want to identify the proper treatment of the right patient at the right time,” said co-horny Dr. Bishoy Morris Faltas, the family of Gellert-John P. Leonard MD Scholar in the field of hematology and medical oncology, as well as the return of medicine and biology of development and development biology at Weill Cornell Medicine, and oncologist at Neyork-Presbyterian/Weill Cornell.

Dr Zilong Bai, Associate in the population of population health sciences, and Dr. Mohamed Osman, an associated PhD student at Weill Cornell Medicine, managed this work together.

Better model, better forecasts

To build a better predictive model, two main scientists have merged. While Dr. Wang’s laboratory focuses on data exploration and the most modern machine learning analyzes, Dr. Faltas is a scientist with knowledge of bubble cancer biology.

They turned to data with SWOG Cancer Research Network, which designs and conducts multicentral clinical trials on adult cancer. In particular, scientists have integrated data from the paintings of prepared tumor samples with the gene expression profile that ensure the shutter of genes that are “enabled” or “off”.

“Since the expression patterns themselves were not enough to predict patients’ answers in previous studies, we decided to get more information for our model,” said Dr. Faltas, who is also a research director at the Englander Institute for Precision Medicine and a member of Sandra and Edward Meyer Cancer Center at Weill Cornell Medicine.

To analyze images, scientists used specialized AI methods called Graph Neural Networks, which capture how cancer cells, immune cells and fibroblasts are organized and interact with the tumor. They also included an automated image analysis to identify these different types of cells at the tumor.

The combination of input data based on images with the data of gene expression to train and test their deep learning model by AI, caused better clinical response forecasts than models used in gene expression or imaging.

“On a scale of 0 to 1, where 1 is perfect, and 0 means that nothing is correct, our multimodal model is approaching 0.8, while unimodal models based only on one data source can reach about 0.6,” said Dr. Wang. “It’s already exciting, but we plan to improve the model for further improvements.”

Searching for biomarkers

When scientists are looking for biomarkers, such as genes that predict clinical results, find tips that make sense. “I saw some genes that I know are biologically important, not just random genes,” said Dr. Faltas. “It was calming and a sign that we were on something important.”

Scientists plan to feed more types of data to the model, such as mutation analysis of tumor DNA, which can be collected in the blood or urine, or spatial analysis, which would allow more accurate identification exactly what types of cells are present in the bladder. “This is one of the key findings of our study-that the data synergias to improve forecasts,” said Dr. Faltas.

The model also suggested several new hypotheses that Dr. Faltas and Dr. Wang are planning to continue testing. For example, the ratio of cancer cells to normal tissue cells, such as fibroblasts, affects the response to chemotherapy forecasts. “Perhaps the abundance of fibroblasts can protect cancer cells against chemotherapeutic drugs or support the growth of cancer cells. I would like to delve into this biology,” he added.

In the meantime, dr. Wang and Faltas will work on verifying their results in other cohorts of clinical trials and are open to the extension of cooperation in order to determine whether their model can predict a therapeutic response in a wider patient population.

The dream is that patients will enter my office, and I could integrate all their data with the AI ​​frame and give them a result, which predicts how they will react to a specific therapy. It will happen. But doctors like me will have to learn to interpret these AI forecasts and know that I can trust them and be able to explain them to my patients in the way they can trust. “

Dr. Bishoy Morris Faltas, Family Gellert – John P. Leonard MD Research Scholar in the field of hematology and medical oncology