K.E. Sanarova1, A.S. Malanova2, D.V. Ryzhkova3, N.S. Goncharova4, O.M. Moiseeva5

1,2,3,4,5 Almazov National Medical Research Centre (St. Petersburg, Russia)

1 St. Petersburg State Electrotechnical University "LETI" (St. Petersburg, Russia)

1 kseniya.sanarova@mail.ru, 2 malanova1998@gmail.com, 3 ryzhkova_dv@almazovcentre.ru, 4 goncharova_ns@almazovcentre.ru, 5 moiseeva_om@almazovcentre.ru

Assessment of lung tissue condition is a key task in the diagnosis and monitoring of diseases associated with impaired structure and function of the pulmonary vascular bed. This problem is addressed using imaging modalities, among which hybrid PET/CT is considered the most informative. However, standard approaches to PET data analysis are largely based on visual assessment and standardized metabolic parameters, which do not always provide sufficient sensitivity to early changes. Manual analysis of PET images is characterized by high labor intensity and interobserver variability, which reduces reproducibility and complicates the use of this method in longitudinal monitoring. This necessitates the development of automated approaches for objective quantitative assessment of changes in lung tissue.

The purpose of the work – development of software for automated quantitative assessment of lung tissue condition based on PET data, designed for the detection and monitoring of pathological changes.

A software tool for automated assessment of lung tissue condition based on PET/CT data has been developed. The algorithm includes sequential steps: conversion of DICOM files into 8-bit grayscale images, spatial synchronization of PET and CT slices based on DICOM coordinates, image preprocessing (filtering, contrast enhancement, inversion), and two-stage segmentation of lung parenchyma and vessels on CT with subsequent transfer of contours to PET images. A function for calculating statistical parameters of intensity distribution in segmented PET regions (mean, maximum, variance, standard deviation) with median aggregation across 10–15 slices has been implemented. A graphical user interface for automated analysis and data export has been developed. The segmentation algorithm was tested on 10 PET/CT studies by comparison with manual annotations performed by a radiologist. Segmentation accuracy according to IoU was 0.87 for lung parenchyma and 0.69 for vessels.

The proposed algorithm can be used in clinical and research settings as a tool for quantitative monitoring of lung tissue condition.

Sanarova K.E., Malanova A.S., Ryzhkova D.V., Goncharova N.S., Moiseeva O.M. Development of software for automated assessment of lung tissue condition in patients based on PET data // Biomedical Radioelectronics. 2026. V. 29. № 6. P. 101−105. DOI: https://doi.org/10.18127/j15604136-202606-17

1. Churko A.A., Nikonova A.I., Halilova L.A., Panina I.Yu., Kulikov A.N. Problemy diagnostiki gipertenzii malogo kruga krovoobrashcheniya. Novye Sankt-Peterburgskie vrachebnye vedomosti. 2024. № (3). S. 56–64. URL: https://doi.org/10.24884/1609-2201-2024-103-3-56-64 (In Russian).
2. Brar K.K. et al. Image segmentation review: Theoretical background and recent advances. Information Fusion. 2025. V. 114. P. 102608.
3. Nguyen D.T. et al. Ct to pet translation: a large-scale dataset and domain-knowledge-guided diffusion approach //2025 IEEE/CVF Winter Conference on Applications of Computer Vision (WACV). IEEE. 2025. P. 1498-1507.