AI could improve prostate cancer diagnosis and treatment
Published Jun 26, 2018
Researchers at KTH and Karolinska Institutet have concluded that AI can contribute to increased understanding of how prostate cancer develops, and even improve clinical diagnosis and treatment of the disease.
Every cancer tumor is unique, with characteristics that change over time. This so-called heterogeneity is due to competing clones within a given tumor, as well as acquired mutations that increase the likelihood of metastases.
Study at the School of Engineering Sciences in Chemistry, Biotechnology and Health
Researchers at Sweden’s
Science for Life Laboratory
have demonstrated how data-driven AI method has the potential to contribute to a better understanding of these major events regarding heterogeneity in prostate tumors and in the surrounding micro environment. The research team from KTH and Karolinska Institutet, led by
, Professor of Molecular Biology at KTH, used data obtained from nearly 6,750 tissue samples with spatial transcriptomics – a method that combines histology (tissue) with quantitative analysis of the active genes, which has been developed by KTH and Karolinska Institutet at SciLifeLab.
The use of spatial information makes a big contribution, Lundeberg says. Analysis of prostate tumor gene activity in a tissue section dramatically increases the granularity, compared to conventional tumor analysis. “We have demonstrated that sampling different parts of the same prostate tumor shows remarkable differences on the gene activity level of the cancer cells at each site as well as the surrounding non-tumor cells, such as cells related to inflammation response likely to be linked to outcome of the patient,” he says.
This rich source of information enables unattended AI methods to identify genetic patterns that cannot be seen by the naked eye, he says. Thus, this massive tissue genetic analysis can serve as a basis for an AI-based clinical evaluation of cancerous tissues and provide insight into gene expression in the tumor's micro environment.
“AI simply helps us to create a computerized tissue anatomy,” he says.
Further insights into the mechanisms underlying cancer are crucial for understanding the progression of tumors and how patients respond to treatment, he says.
Molecular data has been used successfully in the treatment of other forms of epithelial cancers, such as breast cancer. Co-author, Emilie Berglund, a doctoral student at KTH, says that recent studies show that it can help with prostate cancer too. “Early remedy of primary prostate cancer is efficient, however differentiating those that will progress to aggressive cases and who will benefit from what treatment is still problematic,” Berglund says. “We hope that this study makes a significant contribution to these aspects.”
The work was supported by Astra Zeneca and SciLifeLab.
Håkan Soold/David Callahan
Three questions for PhD student Emilie Berglund
PhD student Emilie Berglund is working in research at SciLifeLab and is co-author of the AI study published in Nature Communications. We asked her three questions about the research work she is doing.
How is spatial transcriptomics used effectively in cancer study and potential treatment?
In the last year, much attention has been drawn to the previously unrecognized intra tumor heterogeneity (multiple distinct and spatially separated tumors within one prostate). Such heterogeneity can limit the success of personalized‐medicine strategies if only single tumor biopsy samples are studied during prognosis. We show in this study that this novel technology can be used to assess intra tumor heterogeneity and to determine molecular characteristics of spatially-separated tumor cells within the prostate. Cancer stratification based on molecular signatures has helped the treatment options of other epithelial cancers, such as breast cancer. Conversely, prostate cancer diagnosis is as yet based entirely on histological entities (eg, Gleason score) prostate-specific antigen (PSA) levels and local disease state (TNM), without molecular data. However, the latest studies suggest that prostate cancer progression can be stratified by using molecular signatures. Early remedy of primary prostate cancer is efficient, however differentiating those that will progress to aggressive cases and who is going to benefit from what treatment is still problematic, and we hope that this study makes a significant contribution to these aspects.
Was there any observation that you found particularly interesting in the course of this work? I found it very interesting to study the link between inflammation and cancer, or how does inflammation lead to cancer? We know that tumors hijack inflammation and use it to accelerate the progression towards full-blown cancer. Dampening inflammation or manipulate it will probably make it much harder for the cancer to grow (and promising therapy is starting to appear). We have shown in our study that inflammation and cancer co-exist together and that inflammation seems to exist even before the cancer itself is visible. I am certain that inflammation will play a big role in cancer prevention and treatment in the future.
What will be your next steps in this research? Current pre‐clinical animal models are inadequate to predict how human intra tumor heterogeneity affects the response to treatments. To address this, we want to study responses to treatment directly in cancer patients, using biopsies taken serially before, during and after treatments from the same patient. Almost all men with advanced metastatic prostate cancer (PCa) respond initially to hormonal therapy, while the treatment stops working and resistance emerges after time, leading to a lethal stage of the disease. Analysis of tumor gene expression with aim to target which genes that cause resistance could enhance our understanding of therapeutic failure. So far, there have been very few studies including the sub-clonal architecture of tumors pre- and post- treatment. We hope that this current study could improve the accuracy of personalized medicine.