Why NanoZoomer

Products

Solutions

Case study

Resources

United Kingdom (EN)

Select your region or country.

CYTOREADER

- a new cloud-based AI dual-stain evaluation platform improves accuracy, efficiency and global access to cervical cancer screening

One of the first large end-point controlled deep-learning based AI studies demonstrated substantially improved specificity in cytological cervical cancer screening, while maintaining a high level of sensitivity.

Cervical cancer is still a widespread disease with ~570 000 cases worldwide and a mortality rate of more than 50%. Screening for cervical cancer is one of the largest and most advanced cancer prevention efforts.

Prevention of cervical cancer is currently based on a combination of human papillomavirus (HPV) vaccination and repetitive screening during a woman’s reproductive years. Yet most cervical HPV infections, which cause positive HPV tests, will not result in precancer. The challenge is to identify which women with positive HPV test results are most likely to have precancerous changes in their cervical cells and therefore should have a colposcopy to examine the cervix and take samples for biopsy, or who need immediate treatment.

Cervical cancer screening is currently still based on Pap cytology dating back to 1943. Specially trained cytotechnologists analyse stained slides to look for abnormal cells and identify precancers before they can progress to cancer. However, these tests are time consuming, not very sensitive, and prone to false-positive findings.

An improved new test approach (CINtec^®PLUS) is based on detection of two proteins, p16 ad Ki-67 in the same cell (p16/Ki-67 dual stain), two markers, which are closely linked to cervical carcinogenesis and HPV onco-protein actions. This Dual Staining (DS) has shown greater accuracy for detection of HPV-related precancers compared with Pap cytology. In March 2020, the U.S. Food and Drug Administration (FDA) approved the manual dual-stain cytology test for women who have received a positive result on a primary HPV screening.

Tissue sample

The manual evaluation of cytological slides for patient diagnosis still has a subjective component. In addition, the organisational needs for handling large amounts of slides, archival and quality control remain. Aside from practical challenges, the question emerged whether assisted or automatic slide reading could match or even further improve sensitivity and specificity of the actual diagnoses.

To address those questions, Prof. Niels Grabe and Dr. Bernd Lahrmann of the Steinbeis Transfer Center for Medical Systems Biology (STCMED), Heidelberg, developed the whole-slide imaging and analysis platform CYTOREADER. Its core is based on convolutional neural networks trained to automate dual-stain evaluation. In combination with the imaging capabilities of the Hamamatsu NanoZoomer S360 Digital slide scanner whole-slide scanner, CYTOREADER offers automated scanning functionality of patient slides. Additional applications of the platform include a slide management system, assisted evaluation and second opinion. STCMED’s cloud-based implementation allows easy global access to CYTOREADER’s diagnostic platform through standard internet access only. Furthermore, the CYTOREADER diagnostic platform performs automatic quality checking of samples and lab slide production, Ensuring a high quality lab QC process, In practical application, CYTOREADER uses its automatic analysis capability to present the 30 most pre-cancer likely tiles in a gallery view for assisting experts in a fast evaluation of a patient samples.

Prof. Niels Grabe
Steinbeis Transfer Center for Medical Systems Biology

To assess the analysis qualities of CYTOREADER, one of the largest end-point controlled AI studies on 4,253 patients was launched in collaboration with Dr. Wentzensen from the US National Cancer Institute. The goal was to investigate if a fully automated dual-stain AI analysis could match or exceed the performance of the manual approach of reading the dual-stain assay. CYTOREADER was compared with both, conventional Pap cytology and manual dual-stain testing in epidemiological studies of HPV-positive cervical and anal precancers at Kaiser Permanente Northern California - as one of the largest US managed health care providers - and the University of Oklahoma.

First, the two liquid-based cytology slide types (ThinPrep^®and SurePath™) were digitized with Hamamatsu NanoZoomer-HT, NanoZoomer-XR and NanoZoomer S360 Digital slide scanner. The three large study populations mentioned above were split into training slides and validation slides. After scanning all samples, the deep learning algorithms were trained with slides manually evaluated for DS-positive cells by three observers. CYTOREADER was then used to evaluate the remaining population samples as an independent and blinded evaluation set, only visible to epidemiologists at the NCI.

The virtual slides were divided by the system into tiles to be evaluated, measured and sorted based on their likelihood of containing pre-cancerous or cancerous cervical cells. Tiles above a certain likelihood thereby constituted a positive patient case which would have subsequently been sent to colposcopy.

Results showed that CYTOREADER’s AI could further improve diagnosis quality of the manual DS stain test vs. Pap test. AI-based dual-stain test had a lower rate of positive tests than both Pap cytology and manual dual-stain, with better sensitivity (the ability to correctly identify precancers) and substantially higher specificity (the ability to correctly identify those without precancers) than Pap cytology. AI-based dual-stain reduced referral to colposcopy by about a third compared with Pap (approximately 42% vs. 60%). The testing method was also robust, showing comparable performance in anal cytology.

Virtual Slide, divided in tiles with ranking after analysis

In conclusion the automated CYTOREADER analysis surpassed the performance of the current standard Pap cytology, reducing the number of false positive results and substantially reducing referral to unnecessary colposcopy procedures. The results also support that this test should be further evaluated as an option for anal cancer screening.

CYTOREADER is currently available for selected customers as a research-only product in local, cloud or hybrid installations. Because the manual dual-stain test has only recently received FDA approval for screening of women who have HPV-positive test results, its use is just getting started. With its option to be used as a cloud-based system with ample computational capacity and storage space, its analysis capabilities are principally globally accessible. Thus, it could also provide diagnostic procedures in areas where sufficient personnel, expertise or infrastructure is lacking. Finally, the results demonstrated that the AI increases diagnostic efficiency that will lead to a substantially better healthcare for people.

For further information see:
Wentzensen et al, JNCI (2020) 113(1): djaa066; doi 10.1093/jnci/djaa066
Goodman, JNCI (2020) 113(1): djaa067; doi 10.1093/jnci/djaa067