AI Eyes Preclinical Cancer Imaging

New tools could dramatically change how scientists analyze cancer data

How we image tumors in the smallest creatures has been evolving rapidly since the 1990s, when laboratories began using bioluminescence—the same process fireflies use to produce light—and fluorescence to illuminate how cancers are progressing in mice and rats.

Optical imaging utilizes fluorescent and bioluminescent probes to visualize and quantify molecular and biological events including investigating disease progression, drug development, targeted biodistribution, and gene expression. Bioluminescence imaging has a lower background and increased sensitivity for detection. Bioluminescence imaging has only limited capabilities for imaging multiple molecular events, unlike fluorescence imaging in which a wide array of spectrally distinct fluorescent proteins or dyes offers simultaneous detection of multiple signals.

AI tools, including machine learning, in the imaging space

More recently, preclinical oncology laboratories, including ours, have been exploring the use of artificial intelligence and machine learning (ML) tools.

If you think about how we conventionally use imaging to analyze tumor progression in animals, it is pretty labor-intensive. Cancer cells from a tumor cancer cell line or a mouse tumor cell line are injected subcutaneously or orthotopically into the rodent model. The animals are then imaged over time to see how or if their tumors respond to drug candidates. Lab staff then look at each and every image and enter into the computer what they are looking at. Not surprisingly, the process is labor intensive and vulnerable to human error. Is there a faster way? AI companies believe there are. 

AI refers to creation of machines or tools that can simulate human thinking and behavior. Machine learning is a subset of AI in which machines or tools learn from data to make classifications or prediction either with or without human supervision. The imaging workflow is operated in digitized domains including image acquisition, reconstruction, interpretation, and reporting final data output. 

The superior advantages of AI 

Certain cancer imaging tasks are repetitive, tedious, and burdensome and in this scenario, AI can aid in improving work efficiency, reducing errors, and enhancing diagnostic performance. In cancer imaging, images acquired from tumor models are pre-processed and transformed as inputs to develop ML algorithms and models. ML model or algorithm maps the input imaging data and learns a simple or complex mathematic function that is linked to the target or output, such as a scientific observation. 

We are partnering with Revvity, a life sciences company, to determine feasibility of automatic organ volumetric analysis from 3D ultrasound images of mice using deep learning. It is possible to determine organ volume and weight at endpoint of a preclinical study, but this approach is not conducive to longitudinal tracking. Alternatively, imaging using ultrasound or magnetic resonance can be used to measure organ size noninvasively and repeatedly, but this method requires data to be segmented in 3D. Segmentation of 3D imaging data is time consuming and labor intensive. Additionally, measurements derived from human-annotated data are prone to inter-user variability and user error. In order to address these challenges, an AI model was generated to segment spleens from diverse sets of 3D ultrasound images taken from several in vivo models. 

AI algorithms are also being applied in the in vitro space. Our site in Freiburg, Germany has recently examined the use of a computer visioning platform to track cellular function and dynamics for thousands of cells simultaneously in a PDX model. The data is exported by scientists and uploaded to the Cloud, where an external partner runs the analysis. 

What we have seen so far with these experiments is promising, and in fact the AI technology is now a service we offer to clients for their in vitro studies. With a third of the work done on an oncology mouse typically imaging, next-generation machine learning tools could propel the field forward. 

With that said, the cost of using them is not cheap, which is one reason they are slow to catch on with clients.

Still, with the world gravitating toward alternative methods that accelerate drug discovery and development, improve outcomes, and reduce the use of animals, the synergies being built between AI and preclinical imaging is an exciting development for the field of oncology.

eBook: "Translational Strategies Accelerate Drug Discovery"
This eBook focuses on translational science and discovery partnerships that advance novel therapies through a collective momentum, leveraging efficiencies in drug discovery to accelerate nonclinical research and increase clinical success.
Download the eBook