Accurate cancer diagnosis is comprehensive and almost always requires the availability of a good medical infrastructure with well-connected doctors. While this is often the case in the western world, the situation is different in many developing and newly industrialising countries. The costs of diagnostic tests are also prohibitive in some countries.
Nevertheless, early and reliable diagnosis of malignant diseases is essential in order to reduce the mortality associated with oncological diseases. A team of researchers has therefore developed an algorithm based on artificial intelligence that can deliver significant advances in cancer diagnostics.
The tool is ultimately based on a blood test. A tube is taken from the patient and examined for seven tumour markers. The analysis is inexpensive and easy to perform. The algorithm called OncoSeek then calculates how likely it is that the patient has cancer. The blood results, as well as other demographic markers such as the age or gender of the person affected, are taken into account. The tool also indicates the most likely tissue of origin of the tumour - limited to nine common forms of cancer (breast, rectum/colon, liver, lung, lymphoma, ovary, pancreas, oesophagus and stomach).
To calculate the reliability of the algorithm, blood samples were taken from more than 7,000 people. Of the participants, around 950 had a confirmed case of cancer (note: the blood was taken before starting treatment). A second cohort of around 1,800 test subjects - 1,000 of whom had cancer - was used for further validation.
With the help of OncoSeek, the false positive rate associated with the evaluation of tumour markers was significantly reduced and the specificity of the test was increased to over 90%. The sensitivity for all types of cancer was around 51%, which meant an accuracy of around 84%. Sensitivity was particularly high for pancreatic cancer: over 77%. The algorithm had an accuracy of about 66% for the tissue of origin of the cancer.
OncoSeek significantly improves the predictive power of seven tumour markers - and can even help to identify the tissue of origin. Even if the algorithm is not perfect, especially for the latter, the results still offer clues for further targeted diagnostics. The tool can make a significant contribution to the early detection of tumours, particularly in developing and emerging countries.