Abstract
The demand for precise brain tumor detection and analysis in cancer
diagnosis and treatment has escalated, necessitating extensive and
diverse medical datasets. However, the acquisition of labeled tumor
images is impeded by intricate dissimilarities between tumor and
non-tumor regions, compounded by the diverse spectrum of brain tumor
types. This study addresses a critical research gap by proposing a novel
approach for robust and trustworthy brain tumor MRI image synthesis.
Leveraging label conditional diffusion models, our approach adeptly
captures specific tumor features, resulting in the generation of
high-quality tumor images. Additionally, a trustworthiness control
mechanism, employing evaluation metrics such as Fréchet Inception
Distance (FID) and Inception Score (IS), ensures generated MRI brain
tumor images meet exacting quality, accuracy, and clinical relevance
criteria. Despite potential limitations in sharpness due to resolution
constraints, our framework excels in overcoming challenges posed by
image similarities and variations in brain tumor images. This surpasses
the performance of conditional generative adversarial networks (GANs) by
producing realistic details and textures. This contribution
significantly advances the field of brain tumor image synthesis.