Andrew Goulter, deputy director of the Medical Technology Department of Cambridge Consulting, a subsidiary of Capgemini Inventions, studied how pathology can take advantage of the boom in the use of digital technology. 

Various forms of digital innovation continue to change our lives and reshape the way we work, rest and play, and seem to touch all aspects of human endeavor. Therefore, it seems strange to me that most pathological specimens are still stored on glass slides—even unbelievable. Fundamentally speaking, the process of sampling, staining and inspection has hardly changed in the last century.

Of course, there is another way. Slides can now be captured digitally, and the FDA approved the use of digitized images of the entire slide in 2017 should usher in a paradigm shift in pathological specimen evaluation. But so far, progress has been surprisingly slow. why? Well, the transition from analog pathology to digital pathology seems to require major changes to the processes and workflows involved. It turns out that this is not easy.

The traditional method of analyzing glass hematoxylin and eosin (H&E) slides is time-consuming and labor-intensive. Each slide needs to be individually inspected by a qualified professional. More importantly, the storage of slides requires a lot of physical space, which is usually not available in hospitals. Therefore, leaving the slides and entering the warehouse makes the management and retrieval of specimens difficult and slow. What if the pathologist wants a colleague's opinion on the specimen? Why then, it must be wrapped in foam wrap and delivered to them by express. It's as if the Internet was never invented.

If pathologists themselves were not such scarce resources, this would not be such a serious problem, especially in developing countries. In some parts of Africa, there is only one pathologist for every 1.5 million people. According to statistics from the Chinese Association of Pathologists, there are 20,000 licensed pathologists in China, serving more than 1.4 billion people. The shortage of pathologists, coupled with the huge cost of managing slide specimens, slowed the process of diagnosis and treatment, and had a significant impact on the prognosis of patients.

At Cambridge Consulting, we are studying ways to solve this problem. Artificial intelligence and hardware projects aim to digitize and manage pathological specimens in a smarter and faster way. We believe that the acquisition, management, sharing, and interpretation of pathological information can now — and probably should — be done digitally. What is needed are high-resolution digital images taken from glass slides that are interactive and easy to share.

An obvious starting point is oncology, where the collection and processing of samples is very time-sensitive, and pathologists are under pressure to provide results to clinicians. One aspect of the digitization process that is relatively easy to win is to reduce the cost of the initial image capture system. We are currently developing a low-cost full slice imaging system, using more modern and novel imaging methods, using computational optics to overcome some of the existing hardware obstacles, which will provide multiple readings. This can bring great benefits, especially where pathology is usually costly or simply unavailable. Indeed, just think about the possibilities. Once the slide image is captured digitally, it can be uploaded to the cloud. Pathology services can then be provided remotely online and the results sent back to the local or regional hospital.

But the problem of working hours for pathologists is still a major bottleneck. Whether it is a number or a glass slide, someone must analyze the sample. So far, artificial intelligence has limited impact on pathology, but we believe that artificial intelligence and machine learning have great potential. Once you digitize the image, artificial intelligence can analyze hundreds of data sets in a matter of minutes. Coupled with the evaluation score, the pathologist can only review specimens below the confidence threshold, saving several hours. A pathologist with good artificial intelligence is a powerful proposition.

In our work in this area, we have developed a project called BacillAi. The project combines pathology with artificial intelligence to successfully measure the patient’s tuberculosis infection and monitor disease progression. Proof of principle shows great hope, and external partners are needed to take on this in order to realize its full potential. It has been proven that markers or diseases can be measured, and we would love to see this process integrated into the workflow. Once pathologists and clinicians interact with this information, the patient’s prognosis can be significantly improved.

Interest in computational pathology is increasing, but one of the key obstacles is the poor versatility of the current algorithms that we must use. Many studies in this field have found that algorithms trained on a set of pathological data perform poorly in the face of unfamiliar data. This problem is most likely due to the lack of AI marking data that requires the creation of powerful general algorithms. The problem is that the generation of labeled data is expensive and time-consuming, which creates major problems when developing better algorithms.

We have been investigating some alternative methods recently. One is to use semi-supervised learning. In this regard, artificial intelligence is trained on both small, labeled data sets and larger, unlabeled data sets. This method uses knowledge from labeled data to extract information from unlabeled data. A project we are currently working on focuses on eight types of breast cancer samples, four of which are benign and four are malignant. This allows us to train two models, each with a different level of granularity. One model learns to classify samples as simple malignant or benign, and the other learns to classify them into each of the eight classes. The results are very encouraging. I firmly believe that now is the time for artificial intelligence, machine learning and better hardware to completely change pathology and save countless lives. 

Part of Fast News Group

© Rapid Medtech Communications Ltd. A Rapid News group company. all rights reserved