Overcoming Digital Disparity
A few simple steps can make AI-enabled digital pathology a reality in the developing world
Talat Zehra | | 3 min read | Opinion
The benefits of digital pathology are well known. However, there are obvious challenges hindering implementation – particularly in the developing world – which makes up more than two thirds of the world’s population. Low- and middle-income countries (LMICs) are the hub of world tumor cases and endemic disease. According to GLOBOCAN, the annual incidence of cancer cases in 2020 was 19.3 million (1). It is estimated that these numbers will reach approximately 28.4 million cases in 2040 – a 47 percent rise from 2020.
It’s also well known that there is a significant shortage of pathologists around the globe, but it is the LMICs that suffer most with the burden of disease. This disproportion is rapidly increasing – as is the prevalence of malignant tumors.
Clearly, a great diagnostic dilemma is arising. And it’s clear to me that the adoption of digital techniques in the field of pathology is crucial to improve patient care and management of disease – especially in LMICs.
Below, I highlight some of the key challenges that must be resolved before digital pathology can be fully implemented in the developing world.
Financial constraints are the largest challenge in LMICs. The cost of digital microscopes, whole slide scanners, and AI-based software are beyond the budget of many low resource organizations. And large data sets demand powerful computing and likely access to cloud systems, which also don’t come cheap – especially when considering the need for adequate data security. Clearly, reliable digital technology needs solid IT infrastructure, which is not guaranteed.
Other barriers also exist. Regulatory challenges can pose a problem in some regions because only a few whole slide scanners have FDA approval that can be used for primary diagnosis. And then there are more personal barriers; pathologists are familiar with conventional microscopes, and the switch to digital pathology may take a lot of getting used to.
Finally, more large validation and proof of concept studies are needed before digital pathology can be integrated wholesale.
The challenges do not prevent resource-constrained organizations in LMICs from starting their digital journey. For example, resources from open-source organizations – many of which offer free access to their whole slide image archives – can be used. Alternatively, microscope-connected cameras can easily acquire digital images, negating the need to navigate the technical hurdles of downloading huge data files. Experts can photograph a region of interest for a particular pathology, annotate them, and essentially create a data set using their own patients. These digital snapshots are small compared with whole slide images, whose size is usually in gigabytes. After the slides have been annotated, the images can be used to train automated AI models in image analysis software. Though it’s true that commercially available tools are expensive, open source software options do exist, including QuPath, Orbit, DeepLIIF, and many others.
Recently, we used the open-source software DeepLIIF on Ki-67 immunohistochemistry images. This cloud-native software with user-friendly web interface can quantify Ki-67 positive tumor cells in different tumors and at different magnifications. We used diagnosed cases of breast cancer at 10x resolution, validated the software for Ki-67 quantification, and then compared manual versus automated quantification (see Figure 1 and 2). The consensus was statistically significant. The software was easily compatible with digital images. With the help of this assistive tool, pathologists can perform size gating (to differentiate tumor/stromal cells) and adjust the intensity of positive tumor cells while also selecting – or excluding – regions of interest.
In other words, pathologists can make disease models without the need for high-tech scanners, large hard drives, or high-speed Internet. By making data digital, pathologists can predict the outcome of disease using data science, opening new horizons for precision medicine. The role of technology vendors will be crucial for technical support, but slowly, with these steps, we can achieve full digitization of pathology in the developing world.
Image Credit: CDC Global / flickr.com
- H Sung et al., CA Cancer J Clin, 71, 209 (2021). PMID: 33538338