fbpx

3 reasons to use Deep Learning in Digital Pathology

By News

Deep learning (DL) is a subset of machine learning (ML) which falls under the category of artificial intelligence (AI). The formation of layers is not only pervasive in the nomenclature of these different types of computer intelligence but also in their architecture. DL is based on artificial neural networks, a layered and connected system of algorithms receiving and processing information.

Traditional ML can only be taught information, or say features in images, extracted and coded by a human. DL on the other hand is highly flexible and scalable, it can learn features by itself, making it an extremely powerful tool for image analysis. The Aiforia Platform enables easy access to deep learning and the creation of algorithms for the analysis of any histopathological image; allowing pathologists to speed up their work, make new discoveries, and work with consistency.

  1. Improves accuracy and speed

For the first time ever, deep learning AI models are able to mimic humans in learning to recognize complex visual features in image data. However, DL is faster, often more accurate and can therefore surpass human capability. It can be deployed in a huge variety of applications, from object quantification to tissue classification based on morphology, and give accurate, quantitative information from biological samples. Pathologists can therefore now automate manual and time-consuming image analysis work.

Time is given back to focus on more important tasks: decision-making, collaborating with peers, and saving time away from being hunched over a microscope for hours on end. With the convenience of a portable tablet or other device the pathologist can now automate analysis and review the data and examine the specimen, anywhere, any time.

In one application, an Aiforia deep learning AI model detected and counted all dopaminergic neurons in rodent brain substantia nigra in 5 seconds, compared to the normal 45 minutes with traditional methods. Furthermore, the speed with which this analysis was completed, did not hinder the accuracy, as this was comparable to the StereoInvestigator used previously in the lab.

  1. Enables new discoveries

Deep learning not only can outperform humans in accuracy and speed, it can also find what the human eye sometimes cannot see on its own. Features that are either too small, heterogenous in their expression, or found in a huge quantity spread over a large area are easily recognizable by AI. The process of discovery can be enhanced, especially when considering the development of novel drug molecules, as deep learning excels at detecting subtle differences between study groups.

The smallest of changes, not possible to visualize manually, can be detected by the AI models. Furthermore, you can train them with external ground truths not readily identifiable from the images provided, such as response to treatment. Predictive models can be trained with outcome data. These then learn to extract and visualize the features associated with the outcome.

In one research project AI models were developed with Aiforia Create to enable a pathologist to quantify the number of liver cells positive for a specific biomarker indicating bile duct injury. Without Aiforia this task would have been nearly impossible, requiring multiple, highly-trained pathologists all detecting and quantifying cells in the exact same, uniform manner for several hours a day for many weeks. The increased quantitative and novel data produced by deep learning AI allows pathologists to better assess and determine diagnostic and prognostic indicators and criteria, leading healthcare many steps closer to providing personalized medicine.

  1. Completes tasks consistently

Inter- and intra-observer subjectivity is a critical issue in image analysis across diagnostic settings, many research fields, and many sample types. Whether a specific feature is scored, can vary from one person to another, and even within the same pathologist from one day to the next. Deep learning AI however doesn’t have bad days, it doesn’t get tired, its performance and observation stays consistent. It does not stray away from what it has been taught to do or find. The algorithms classify results and solve problems with concordance, always according to the ground truth they were given.

In one such application, Aiforia was deployed by pathologists at the pharmaceutical company Sanofi to count large quantities of Th+ neurons. Over 150,000 objects were detected in the samples for this one study. Not only was the process speedier than with traditional, manual methods but absolute consistency was achieved. The AI models analyzed and quantified each slide and section with the exact same ground truth in mind. Find out more about how Aiforia enabled this preclinical research.

Conclusion

Aiforia is a software platform allowing users to create their own deep learning AI models to automate the analysis of a variety of tasks in a variety of histopathological images, without any coding or hardware needed. Deep learning AI should be viewed as an assistant to the pathologist, not a replacement. A tireless, consistent, speedy assistant, here to help you discover more. Find out more about the Aiforia Platform here.

This article originally appeared on www.aiforia.com, read the original article here: https://www.aiforia.com/blog/3-reasons-to-use-deep-learning-in-digital-pathology/

The first aiForward project commences

By News

“I have been thinking about these vessels, a lot. I am excited to finally be able to see them in a tissue context.”

Reidunn Edelmann, an MD currently doing postdoctoral research at the University of Bergen at the Centre for Cancer Biomarkers (CCBIO) in Norway, is one of the first successful aiForward applicants. Driven by a passion for clinical research, Reidunn completed his PhD in cell biology after her medical studies and is currently focusing on studying biomarkers and vessel characteristics in cancer. She is interested in changing paradigms over how biomarkers and vessels are studied, from simple “very much just focused on one marker,” as she explains, “to thinking about the wider, more functional picture.”

Deepening cancer research

The vasculature, the arrangement of blood vessels in a tumor, is a significant component of analyzing tumor microenvironment which in turn is considered an important prognostic factor in cancer. “Earlier analyses of tumor vasculature in clinical samples has generally been of “low-content” nature,” Reidunn states. Her mission is to take the analysis of this important biology to the next level by deepening the characterization of these vessels and to ultimately correlate them to patient outcome-related end-points as well as other biological traits of tumors in different cancer types.

In order to fulfill these goals, Reidunn and her lab created a complex, multiplexed fluorescent staining that potentially can visualize distinct vessel phenotypes, based on activation of an essential fate-decisive pathway. However, the next step proved to be difficult, “The samples are so heterogenous, the patterns are so different from each other even for the same staining quality,” Reidunn explains “Current and traditional methods to analyze these were too simplistic, not suitable for this complex staining.” She had the samples, she had the staining, but she did not have the necessary tools to enable analysis and discovery.

aiForward enables analysis

Reidunn came across Aiforia at a conference, intrigued by the possibilities of applying deep learning AI to her research she contacted one of Aiforia’s scientists to describe her problems: “Manually it would definitely not be possible to work on my research, it takes too much time with these complex samples.” She was encouraged to apply to the aiForward program, as she explained: “My motivation is that I really, really want to analyze complex, high-content samples. Deep learning AI can learn the different patterns of what the activation of one pathway looks like.”

Her project with aiForward has started off with uploading the samples of her cohort, 235 colorectal cancer samples of patients to the Aiforia Cloud Platform — the Aiforia scientists were actually able to build a custom solution to support the upload of her unique sample set. When asked how she feels about the next steps, which involves training the AI model to recognize her features of interest, Reidunn excitedly answers: “I think it will be so easy, Aiforia (software) does all the hard work, I don’t need to think about any algorithms. All I need to do is see my slides and look for my feature and annotate it and then the program is going to do the rest. That’s the amazing thing. I myself am not so into digital, technical stuff. I am very much biology focused. So, thanks to Aiforia, I can just focus on what I want to focus on, the biology.”

Applications to the aiForward program are accepted on a continuous basis.

Apply Now