1 results listed
The accurate and timely diagnosis of renal
pathologies, including cystic lesions, stones, and tumors, remains
a critical challenge in clinical practice. Traditional manual
analysis of computed tomography (CT) images is inherently
susceptible to inter-observer variability and is resource-intensive,
necessitating the development of robust automated diagnostic
methodologies. This study introduces an artificial intelligencedriven
framework, employing the state-of-the-art YOLOv11 deep
learning architecture, for the classification and segmentation of
renal abnormalities within CT dataset. Utilizing the publicly
available CT KIDNEY DATASET: Normal-Cyst-Tumor and
Stone, comprising 12,446 unique CT slices (3,709 cyst, 5,077
normal, 1,377 stone, and 2,283 tumor), a multi-stage pipeline was
developed. Initially, an expert reader performed rigorous
segmentation of kidneys, stones and cysts excluding extraneous
anatomical structures. Subsequently, to enhance model
robustness and generalization, a suite of preprocessing
techniques, including median filtering, contrast-limited adaptive
histogram equalization, and comprehensive data augmentation,
were implemented. A YOLOv11 classification model was then
trained to accurately distinguish between normal kidneys and
those affected by cystic lesions, stones, and tumors, achieving an
F1-score of 0.9993 across all four classes. Following the
classification, dedicated YOLOv11 segmentation models were
trained, utilizing the classification results for normal, stone and
cyst classes, to achieve precise delineation. The performance of
the segmentation models, measured by mean Average Precision,
yielded scores of 0.9951 for kidney, 0.9670 for stone, and 0.9522
for cyst segmentation. A comprehensive experimental study has
been conducted, and the proposed model has achieved high
performance for both classification and segmentation tasks. In
conclusion, this automated system can be used as a decision
support tool for radiologists, potentially shortening diagnostic
delays and improving overall diagnostic accuracy.
International Conference on Advanced Technologies, Computer Engineering and Science
ICATCES
Abdullah Aldemir
Huseyin Kocaturk
Selcuk Levent Gorgec
Selcan Kaplan Berkaya