The role of computed tomographic sarcometry data using machine learning technologies in predicting postoperative outcomes in patients with gastric cancer

Background. Sarcopenia is a negative prognostic factor in cancer patients. This is important in patients at high risk of developing nutritional deficiency. Determination of the skeletal muscle index (SMI) with the help of computed tomography (CT) the method of choice to is diagnostics of sarcopenia. However, the clinical use of CT is limited by the increased time required to manually measure muscle mass from CT-images.

Aim. To improve the use of CT sarcometry at the preoperative stage of combined treatment in patients with gastric cancer to stratify the risk of postoperative complications using the developed software assistant.

Materials and methods. At the first stage, a “dataset” was created. It contained 207 CT images. It was used to train a muscle tissue segmentation model. The Dice’s similarity coefficient was achieved at a value of 0.91 on a small training set. At the second stage of the study analyzed the incidence of sarcopenia in the examined patients before neoadjuvant chemotherapy and immediately before gastrectomy; 41 (63%) of 65 patients had sarcopenia in the study group and in 50 (77%) patients after neoadjuvant chemotherapy. Postoperative complications were diagnosed in 12 (19%) of 65 patients. There was no correlation between the frequency of their occurrence and the muscular status of patients (p=0.392), however severe complications (≥IIIb according to the Clavien–Dindo classification) were detected only in the group of patients with sarcopenia (p<0.001).

Results. As a result, preoperative sarcopenia is a negative factor in the development of severe postoperative complications in patients with gastric cancer who have undergone gastrectomy. The introduction of deep learning technologie to clinical practice can facilitate the assessment of muscle tissue parameters in patients with cancer.