Introduction: Segmentation is the process of delineating tumors and critical organs in
medical imaging and is essential for further clinical assessment and patient treatment.
Computed tomography (CT) and positron emission tomography (PET) play a crucial role in
tumor analysis, providing valuable information about tumor location, anatomy, and disease
stage. Traditionally, segmentation is still performed manually by physicians on PET/CT
images; however, it is time-consuming and labor-intensive. In recent years, the development
of automatic segmentation methods based on deep learning has emerged, offering a potential
alternative to manual segmentation. Purpose: The goal of this master's thesis is to present
the use of automatic segmentation methods for segmenting tumors and organs on PET/CT
images and with a systematic literature review explore significance of use from the
perspectives of both healthcare professionals and patients. Additionally, the thesis focuses
on identifying various automatic segmentation methods, comparing their segmentation
accuracy, time requirements, and clinical applicability, as well as examining the challenges
associated with implementing such approaches into clinical practice. Methods: In this thesis,
descriptive and systematic literature review methods were used. The literature search was
conducted across multiple databases using inclusion and exclusion criteria along with
relevant keywords. From an initial 1,393 search results, 34 most thematically relevant
articles were selected and included in the thesis. Results: We presented the results in two
tables, in the first table, we introduced different models of automatic segmentation and
compared their effectiveness, the use of studies from various datasets, and their clinical
applicability, while in the second table, we focused on the challenges that arise when
implementing such methods in a clinical setting. Discussion and conclusion: Automatic
segmentation methods enable faster, more reproducible, and objective segmentation of
tumors and organs compared to manual segmentation. This contributes to the advancement
of computer-aided diagnosis and treatment planning, enabling the identification of important
prognostic factors and monitoring of treatment response. Despite numerous advantages,
these methods have not yet been widely implemented in clinical practice due to technical,
legal, ethical, and privacy-related challenges. With the collaboration of academia, industry,
and clinical experts, it will be possible to implement these methods into practice and
accelerate diagnostic processes and treatment planning, which is vital for optimal patient
care and reducing the burden on healthcare professionals.
|
