HomePortfolioDeveloping a Method to Improve the Accuracy of Melanoma Stage Diagnostics

Developing a Method to Improve the Accuracy of Melanoma Stage Diagnostics

Healthcare
AI / ML

The National Scientific and Practical Center of Oncology and Medical Radiology asked to develop highly sensitive statistical decision procedures for diagnosis of metastatic lesions in regional lymphatic nodes.

Developing a Method to Improve the Accuracy of Melanoma Stage Diagnostics

The customer

The customer is the National Scientific and Practical Center of Oncology and Medical Radiology–a leading oncological institution in Belarus. The center employs 1724 people, including 323 doctors who, in total, hold 22 Ph.D., 62 Candidate of Medical and Biological Sciences, and 10 professorial degrees. The clinic of the center can accommodate up to 820 patients. Every year it provides medical examination services as well as treatment to over 18 thousand people suffering from cancer.

The need

Melanoma of the skin is one of the most common types of malignant tumors. The standardized level of incidence for European population increases twofold every 10-20 years (the number of occurrences increases from 3% to 7% annually).

At the moment, the standard treatment for those suffering from melanoma is a surgery, which, in many cases, involves removal of regional lymphatic nodes (the procedure is also known as regional lymphadenectomy). However, the results of a long-running research by an international group of experts in diagnostics and treatment of melanoma supported by the World Health Organization showed that regional lymphadenectomy should only be prescribed in case there are metastases in the lymphatic nodes (metastatic lesions of regional lymphatic nodes). In addition, after a surgery, a patient has to spend much more time in hospital–this is another argument for abandoning the procedure of lymphadenectomy as a preventive measure. Apart from that, the resulting post-operational complications can lead to disabilities in people who have been cured from melanoma.

Thus, it is very important to single out only those people with melanoma who really need to be treated with preventive regional lymphadenectomy. Before we started our work, the accuracy of diagnosis based on the previous research and mathematical methods designed to determine whether regional lymphadenectomy should be prescribed to certain patients did not exceed 58.2%.

So, the task was to develop highly sensitive (accurate) statistical decision procedures for diagnosis of metastatic lesions in regional lymphatic nodes based on indicators that characterize primary tumors and the patient’s organism in general.

The solution

We used a generalized discriminant analysis model to create statistical decision procedures for diagnosis. To increase diagnostic efficiency of the decision procedure we developed based on a generalized discriminant analysis model we decided to select a group of patients who needed additional methods for examination of the regional lymphatic system in order to tell, if there were any metastatic lesions. In this case, the training sample consisted of two groups of people under examination and the decision procedure of the classification classified the patients suffering from melanoma of the skin into three groups:

  • those who required regional lymphadenectomy
  • those who did not require this procedure
  • those who were to be sent for further examination in order to determine their status more accurately

As the result, we developed a software system that can diagnose metastatic lesions of regional lymphatic nodes in people suffering from melanoma of the skin.

The outcome

With the statistical methods developed by our team we could achieve diagnosis accuracy of over 83% for patients with metastatic lesions of regional lymphatic nodes. A software system for diagnosis of metastatic lesions of regional lymphatic glands was developed and introduced in oncologic dispensaries of Belarus. This decreased the number of surgeries and–consequently–the incidence of post-operational complications and helped to save the lives of hundreds of patients.

/

Want to develop something similar?

Preloader
Yauheni Starchak

Yauheni Starchak

Artificial Intelligence Practice Head

y.starchak@altoros.com650 265-2266

4900 Hopyard Rd. Suite 100 Pleasanton, CA 94588