Skip to main content
European Commission logo print header

Screening of Gastric Cancer via Breath volatile organic compounds by Hybrid Sensing Approach

Periodic Reporting for period 3 - VOGAS (Screening of Gastric Cancer via Breath volatile organic compounds by Hybrid Sensing Approach)

Reporting period: 2022-01-01 to 2022-09-30

Screening for different diseases is required to reveal groups of individuals in risk of a specific disease and who could benefit from early detection. The ideal screening test is low-cost, high-accuracy, non-invasive, easily repeatable, and practical. In the VOGAS project, we addressed these requirements by integrating different nanotechnologies into a single device to detect exhaled breath. This approach relies on convergence of three different sensing technologies, which are presumably orthogonal in nature. A single breath sample was analyzed by Gold Nanoparticles sensor array, MOX sensor array and a Hollow Waveguide IR spectrometer, all in parallel in a matter of minutes. This device was placed in five different clinical sites, in different countries and continents. The sensor response was recorded, stored and pre-processed and sent to the VOGAS cloud for elaborated statistical pattern recognition analysis. In general, the different statistical programs compare the responsive pattern of the sensor array and finds the features that characterize the gastric cancer patients from healthy samples. This analysis is then translated to a "screening result" harboring a level of certainty of the particular breath sample, originating from a "Sick" or "Healthy" individual. If we look at each of the units as a separate device, we have achieved about 80% accuracy of the screening results, however, if we combine all units and force the algorithm to use the same parameters, we only reach about 74% accuracy. This is to be expected since the units are never made exactly the same and site to site, as well as geographical differences can play a part. In fact, this is one of the reasons that the Vogas project was planned to take place in so many different locations.
The project was extended in 9 months due to delays caused by the covid 19 pandemic. However, we still managed to perform all of the planned activities.
Briefly:
• The Sensors array was further optimized, and sensor redundancy was adopted from a research group of Prof Denatale in Rome who demonstrated a significant noise reduction and drift compensation. In the last year of the project, we initiated a collaborative work with them for the analysis of the Vogas results. It seems that the implementation of the "adaptive neural network method" (ANN) developed by the Denatale group, works well on the Vogas results.
• Device manufacturing - Due to the need to test the VOGAS device in clinical trials, including a large number of subjects in a limited time, it was decided to build six devices of the final format, instead of three which were originally planned. This decision was discovered to be of great importance due to the outbreak of covid-19 which delayed some of the activities, limiting the actual time to recruit GC patients. Thus, six devices were built, distributed and collected breath samples for WP4 and WP5 at the different clinical centers.
• VOC analysis – UOL, PUC, NCIU, ACCCC, and CJO-HUSI collected samples of gastric cancer tissue and stomach content form GC patients as well as from heathy volunteers. These were shipped to UIBK for Volatile Organic Compound (VOC) analysis by GC/MS. The analysis was done along with a tissue culture study comparing VOC emission from normal and gastric cancer cell lines. This study comprising most of the work of work package 1, led to the finding of several unique and novel VOC’s whose levels are altered in malignant tissue relative to normal ones. These findings are currently submitted to scientific publication.
• Clinical studies - All five clinical centers, obtained the necessary permits, agreed upon Standard Operating Procedures (SOPs) and the two main clinical studies were launched (WP4 and WP5). Two units were sent to Latvia, one to Brazil, one to Chile and one unit to Ukraine. Later, two of the units (from Chile and Ukraine) were fixed, re-calibrated and sent to Colombia. We have put the two Colombian units in parallel so we could compare their function on the same patients. Despite some technical problems we recruited all of the planned number of subjects for these studies.
• Obtainment of results – JLM has dedicated a very large storage space in the Vogas cloud. Each center uploaded the Vogas breath sample results to the cloud where it was stored for statistical analysis. Both Juha and Johannes inspect the results daily to make sure that both the GNP and the IR units are generating meaningful results. This proved to be very important since technical problems did accrue, and we needed to respond in a timely manner.
• Statistical analysis – Breath samples of healthy volunteers and gastric cancer patients were analyzed by VTT according to conventional statistical methods. The results show about 80% accuracy for single clinical site. This is an average accuracy figure coming from four units, after excluding technical problems. However, when we force the model to use the same parameters for all units, the accuracy level drops to about 74% of the combined generic model of all sites. In order to increase our analytical capability, we have initiated collaboration with the Denatale group in Rome. This group is the inventor and world leader in what is called "adaptive neural network method" (ANN). We have sheard a partial set of results with the Denatale group and their initial analysis shows about 88% accuracy for a single site. We will follow up with the complete set of results and tentatively add this to the final publication.
• Ethics – An external ethical board (EEB) was formed, and all informed consent forms were evaluated. Periodic ethical meetings were held, and the ethical summaries of these meetings were included into the periodic reports. Finally, a responsible research and innovation (RRI) study and a few filed exercises were performed.
The existing screening methods for cancer suffer from a multitude of disadvantages. Some, such as colonoscopy and sigmoidoscopy are invasive and hence, pose actual risks of medical complications to the screened subjects. Others, such as TEM or mammography for example, require special facilities, are expensive and subject the examined individual to risks of radiation. These caveats are compounded by the discomfort and inconvenience, as well as the expenses of time and money these screening methods impose on the patient. These are all good explanations for the low number of screening methods implemented today. Despite the above, it is widely agreed that one of the most curtail obstacles in screening programs is that a very limited number of diseases currently have an effective screening approach available, e.g. screening for only four cancer types is currently recommended in the EU. The VOGAS project is based on three different and most likely orthogonal analytical methods. Therefore, while maintaining the ease of operation and un invasiveness, VOGAS has a very good chance for improving signal to noise ratio and hence it’s predictive value. We currently demonstrate about 74% accuracy for distinction between cancer and control for a cross site model. This is based only of the GNP and MOX sensors without the aid of the IR analysis. In light of the above, we believe that this approach holds great potential and great promise for the future of disease screening and early detection.

Related documents