Contrary to popular belief, prostate cancer is most common in men, followed by lung cancer and colorectal cancer. In France, more than 40,000 cases (of prostate cancer) are diagnosed each year . The invasive nature of the most effective means of diagnosis (a biopsy) is one of the causes of late management of this type of cancer, dramatically increasing the risk of death - currently estimated at 8,000 cases per year. Recently, a team of Korean researchers developed a very efficient diagnostic method based on artificial intelligence , which is believed to be close to 100% reliable.
When there is suspicion of prostate cancer, the doctor begins by recording the level of PSA (Specific Prostate Antigen), a cancer factor used as a clue and present in the blood. However, since the accuracy of this first diagnostic means is only 30%, a considerable number of patients undergo an additional invasive biopsy to confirm it. Most therefore suffer from the resulting side effects, such as bleeding and pain.
Recently, the Korea Institute of Science and Technology (KIST) in South Korea announced that the research team led by Dr Kwan Hyi Lee from Biomaterials Research Center and Professor In Gab Jeong from Asan Medical Center has developed a technique that can diagnose prostate cancer from urine in just twenty minutes, with nearly 100% accuracy. The research team developed this technique by introducing an intelligent analysis method based on artificial intelligence (AI), using a biosensor that is highly sensitive to electrical signals. The details were published in the journal ACS Nano .
A diagnostic method based on several cancer factors
As a non-invasive method, a urine-based diagnostic test is convenient for patients and does not require a biopsy, allowing cancer to be found without side effects. However, because the concentration of cancer factors is low in urine, urine-based biosensors have so far been used only to classify risk groups rather than to make an accurate diagnosis.
To alleviate the problem, Dr Lee's team at KIST is developing a new technique for diagnosing the disease from urine, but using a highly sensitive biosensor based on electrical signals. One of the approaches they tested, using a single cancer factor, achieved diagnostic accuracy of over 90%. However, to overcome this limitation, the team simultaneously used different types of cancer factors instead of using just one. This has allowed them to improve diagnostic accuracy in innovative ways and achieve amazing accuracy.
Almost 100% accuracy thanks to the power of AI
The team has developed an ultra-sensitive solid-state sensor system capable of simultaneously measuring traces of four selected cancer factors in urine to diagnose prostate cancer. They then trained the AI using the correlations between the four cancer factors, which were obtained from the newly developed sensor. The AI algorithm thus formed was used to identify people with prostate cancer by analyzing the complex patterns of the signals detected. The diagnosis by IA analysis allowed the classification of 76 urine samples with an accuracy of almost 100% (99%).
The 76 urine samples were measured three times in all, thus generating 912 biomarker signals, or 228 sets of detection signals. The researchers then used two types of algorithms - RF (decision tree forests) and NN (k nearest neighbors method) - to analyze the signals from the multi-markers. Both types of algorithms resulted in increased precision as the number of biomarkers increased. © Korea Institute of Science and Technology (KIST)
“ For patients who need surgery and / or treatment, cancer will be diagnosed with great accuracy using urine to minimize biopsies and unnecessary treatments, which can significantly reduce medical costs and provide relief to medical personnel. Said Professor Jeong of Asan Medical Center. “ This research has led to the development of an intelligent biosensor that can quickly diagnose prostate cancer with almost 100% accuracy, using only a urine test. And it can be used to accurately diagnose other cancers using a urine test , ”adds Lee.
In the near future, it is highly likely that we can benefit from the prowess of machine learning systems to detect many other forms of cancer and diseases that require early diagnosis for effective management. Combined with new treatments currently in development, overall survival rates after cancer diagnosis are expected to increase sharply in the decades to come.