How Computer Vision Enhances Cancer Research

Computer vision is artificial intelligence that allows algorithms to extract meaningful information from videos and images. Cancer researchers have explored effective ways to use it to examine pictures, microscopic samples, medical scans and more. Some approaches can shorten previously cumbersome workflows, allowing resource-stretched teams to achieve goals and heighten patient impacts.

Improving Knowledge of Tumor Growth Drivers

After confirming the presence and type of cancer in biopsies, pathologists may perform genetic sequencing on the RNA molecules within the samples. Then, they can find which genomic changes influence the tumor’s growth. That information furthers valuable research and personalized interventions. However, current methods’ costliness and lengthy processes leave some researchers eager for viable alternatives.

One team built an AI tool that analyzes standard microscopy images of biopsies to predict the genetic activity within tumor cells. They trained their innovation on over 7,500 samples representing 16 cancer types and other relevant datasets, including pictures of healthy cells.

These researchers prioritized usability through easy interpretability, creating their AI-powered program to show the gene-related information as a visual tumor biopsy map. That decision allows users to identify distinctive variations in specific areas. The group also relied on a standard staining method to visualize cancer cells, and the tool identified the genetic expressions of more than 15,000 genes within the stained images.

Their findings indicated a correlation of over 80% between the AI-predicted genetic activity and actual behavior. The model generally performed better when the sample dataset included more examples of a specific cancer type.

This research team’s experiments also showed the algorithm’s potential validity for assigning genomic risk scores to patients with breast cancer. Parties categorized as riskier had more recurrences and shorter durations between them.

People have used AI for other fascinating medical advancements. One development can detect COVID-19 with up to 99% accuracy, showcasing an essential public health enhancement. Despite the impressiveness of these possibilities, professionals must only complement their work with them. Letting AI replace firsthand experience could reduce positive patient outcomes.

Finding the Most Appropriate Treatments

People undergoing cancer-related interventions detail the stress and unpleasant symptoms associated with potentially suboptimal solutions. Although many individuals tolerate nausea, hair loss and more, they become less agreeable to continue if early tests do not show promising results.

Everyone benefits if cancer specialists identify the best patient-specific treatments sooner. The typical approach to designing care plans involves studying CT and MRI scans with only one data point per pixel, represented as shades of gray.  Some researchers use AI to make headway. One tool can examine up to 30,000 details per pixel and analyze tissue samples as small as 400 square micrometers — about the width of five human hairs.

The team used donated samples to assess the outcomes. When applied to bladder cancer cases, the AI platform found a specialized cell group that creates tertiary lymphoid structures. Current knowledge suggests that these improve patients’ immunotherapy responses. Additionally, the tool differentiated between cancerous cells and tissue mucosa in gastric cancer samples, helping users more accurately pinpoint the extent of their spread.

These researchers believe their efforts could show doctors which treatments work best for various cancers. If so, it may also streamline relevant research by helping them extract more valuable data from common diagnostic images.

Shortening Drug Development Timelines

Making new cancer treatments commercially available takes years, and the prospects hinge on successful clinical trials. Researchers in London recently created an AI-enabled approach to studying how well drugs reach their targets. Focusing on the most effective options could improve outcomes, convincing regulators to widen product availability.

The group used almost 100,000 3D microscopy images of melanoma cells, and geometric deep learning algorithms analyzed their shape. Previous efforts only obtained two-dimensional data from samples on microscope slides, but this approach examines cells as they appear in bodies. Moreover, it reveals how they change shapes due to specific treatments and shows variability across cellular populations.

This tool was more than 99% accurate in detecting how specific drugs affected the cells. It even identified shape changes triggered by those targeting different proteins.

Because the AI revealed biochemical alterations, the researchers think their innovation could highlight particular targets to emphasize with new cancer medicines. Then, the software would shrink the preclinical time frame from three years to three months. Relatedly, it could reduce trials by up to six years, more quickly finding the patients most likely to benefit and pinpointing the common side effects.

Streamlining Cancer Evaluation Tasks

AI has already enhanced cancer researchers’ duties, but most tools only handle individual parts of the workflow. That means medical specialists interested in integrating the technology into their workdays need to learn to use several products. However, some groups want to build multipurpose solutions to increase user-friendliness.

One built a model similar to ChatGPT. They used it for multiple evaluative processes linked to 19 cancer types, showing its versatility. More specifically, it accelerated evaluation tasks for enhanced detection, prognosis and treatment responses. The developers also believe their innovation is the first to predict and validate outcomes across several international patient groups.

The AI model reads digital slides containing tumor samples, analyzes the molecular profiles and finds cancerous cells. It also examined the tissues surrounding growths, which indicate how well patients have responded to standard treatments or show researchers which are less effective. Experiments suggested it was more accurate than currently available products. Additionally, it linked particular tumor characteristics to increased patient survival rates for the first time, potentially unlocking new research areas.

The researchers trained the model on 15 million unlabeled images split into chunks depending on areas of interest. A later step exposed the algorithms to 60,000 whole-slide examples representing the 19 cancer types. This approach taught the AI to assess entire pictures for thorough results.

Then, the group tested their tool on 19,400 whole-slide images found in 32 independent datasets. Because that information came from 24 globally located patient cohorts and hospitals, it provides an accurate sample of real-life conditions.

Enhancing the Value of Biomedical Microscopy Images

Cancer researchers use biomedical microscopy images to further their work, but existing workflows take days to examine this data. One team developed a new computer vision technique to make those essential tasks more efficient. It uses machine learning to analyze samples and find shared characteristics among cancerous tumors.

The tool efficiently obtains results by examining multiple areas of individual growth and perceiving them as a whole. Other products that analyze biomedical microscopy images divide large tumors into smaller patches and treat the portions as separate samples. However, these pictures can contain up to 1 billion pixels, so they are time-consuming to study.

The developers envision clinicians could make near-immediate diagnoses from tumor images. Then, those professionals would pass the information to surgeons performing operations to extract cancerous tissues, allowing them to use the most current insights.

Tests comparing this tool to the best-performing baseline image analysis techniques showed it was almost 4% better and achieved nearly 88% accuracy in some cases. The researchers also stressed that users could apply it to any tumor type and microscopy method, making it broadly applicable.

Pushing Cancer Research Forward With Computer Vision

AI-driven computer vision can elevate cancer researchers’ output, maximizing their scientific and patient-related results. These examples illustrate the abundant potential, but professionals interested in applying the technology should do so to augment acquired expertise and not treat innovations as foolproof.