Steinbeis experts tap into artificial intelligence for the early recognition of cervical cancer
Cervical carcinoma – more commonly known as cervical cancer – does not normally start with any obvious symptoms, or discomfort that makes it easier to make an early diagnosis. It can, however, be spotted relatively easily by conducting early tests. The Steinbeis Transfer Center for Medical Systems Biology (MSB) has developed a cloud-based artificial intelligence platform called CYTOREADER, that helps labs introduce and run improved cervical cancer screening using a method called dual-stain cytology (CINtec® Plus). As well as improving the accuracy of diagnostic testing, it also enhances the efficiency of screening.
The aim of cervical carcinoma screening is to detect precancerous lesions so that the condition can be treated, thus preventing the development of invasive carcinomas. There have been important advances in the prevention of cervical carcinoma in recent decades, also made possible by Harald zur Hausen’s discovery that the cause of the disease is the human papilloma virus (HPV). Cervical cancer can therefore by prevented by a combination of vaccination and regular screening tests.
Early detection tests using the dual-stain test
Cytological detection of altered cells from cervical smears is central to the use of screening programs to drastically reduce cervical cancer mortality in developed countries. The basis for this came as early as 1943, when experts starting using so-called Pap tests. Evaluating Pap tests was difficult, however, time-consuming, not particularly accurate, and had a tendency to produce false positives. In recent years, attention has therefore turned to dual-stain testing, which was developed for the use of biomarkers in cytological testing. Testing can detect the simultaneous expression of two proteins in cells: p16 and Ki-67. These indicate raised levels of cell division (Ki-67) coinciding with a malfunction of cell division (p16). In March of this year, the testing method was approved by the United States Federal Food and Drug Administration (FDA), resulting in improved early detection testing. Although manual assessment of dual-stain tests is more consistent, specific, and sensitive than with Pap tests, to a certain degree it is still susceptible to subjective opinion. Glass slides containing cellular material of the cervix are examined cytologically under the microscope for the presence of the two proteins.
Automating screening using CYTOREADER
Experts at the Heidelberg-based Steinbeis Transfer Center for Medical Systems Biology have therefore developed the CYTOREADER platform, which automates this final subjective stage of cervical screening by using artificial intelligence. Their new platform has been comprehensively assessed in initial epidemiological studies in partnership with the American National Cancer Institute and the health company Kaiser Permanente in northern California. The studies involved 4,253 patients (Journal of the National Cancer Institute, June 25, 2020). Compared to standard procedures (Pap cytology), the studies were able to reduce colposcopies (biopsies) by 30%, without compromising the detection of precancerous lesions. CYTOREADER operates fully automatically in the background and can assess the quality of microscopic samples and support diagnostic decision-making. The AI system makes it possible to improve the quality of diagnosis by enhancing sensitivity and specificity, also making the screening program more efficient. Using the fully automated slide scanner allows cytological slides to be digitalized fully automatically with microscopic definition.
To do this, complex image processing techniques are used, such as deep learning networks, which can be trained by sharing examples of good and bad quality. The digital images of patient samples are then uploaded to the cloud where they are processed, assessed, and archived. The cloud system can be accessed via standard browsers, making it possible for experts anywhere in the world to analyze samples via the internet.
These fully automated deep learning networks for analyzing cellular cytological samples are the backbone of the CYTOREADER system. They were trained by using samples of two types of thin-layer cytological slide preparations (ThinPrep® and SurePath™). To conduct its analysis, CYTOREADER breaks down images of digitalized slides into thousands of tiles, which are sorted according to the level of cancer risk. Physicians or cytologists making a diagnosis are then shown a gallery of the 30 most pronounced examples of cancer precursors. In the studies, it took less than a minute and only a few clicks of the mouse or keyboard to make a diagnosis.
Improved diagnosis and greater capacity
Direct comparisons with Pap cytology showed that CYTOREADER can significantly improve the already enhanced diagnostic quality of dual-stain testing. The number of inaccurate positive diagnoses dropped significantly (due to the higher specificity) and the detection of actual cases improved (due to higher sensitivity). The number of positive patients recommended for colposcopy (invasive tissue biopsy) could be reduced by 60% to only 42% of HPV-positive cases. CYTOREADER thus significantly outperforms the current performance standard of Pap cytology in diagnostic terms.
To use CYTOREADER, a slide scanner is needed. This can also be set up as a service in a local laboratory. Sample logistics also have to be adapted, but on the other hand, using the cloud offers clear advantages in the course of projects. Laboratories no longer require complex, maintenance-intensive, in-house IT systems. In addition, patient samples can be made accessible anywhere in the world 24/7. The evaluation capacity provided by the computer system for deep-learning networks is almost infinitely scalable, simply by bringing in extra cloud computing resources. Digitalizing labs thus makes it possible to reorganize certain stages of the value chain. This will play an essential role in driving change in the business landscape of laboratories in the coming years, and it is entirely possible that additional market concentration will take place. Overall, the cloud has the ability to sweep local technological hurdles aside, allowing lab services and experts to focus on their core competences.
Breaking down value chains in this way, even beyond regional and national boundaries, will also have a significant impact on global practices. Since 80% of cervical cancer cases occur in developing countries and emerging economies, the cloud has the ability to catapult such countries into state-of-the-art medical technology. Subsequent to the positive study findings, the plan now is to submit CYTOREADER for FDA or IVD approval.
More information on CYTOREADER: www.cytoreader.com
Prof. Dr. Niels Grabe (author)
Steinbeis Transfer Center Medical Systems Biology (MSB) (Heidelberg)
Dr. Bernd Lahrmann (co-author)
Steinbeis Transfer Center Medical Systems Biology (MSB) (Heidelberg)