Automated diagnosis of acute cerebral ischemic stroke lesions from diffusion-weighted magnetic resonance images in assessment with modified Rankin Scale
Affiliations: [a]
Department of Electronics and Communication Engineering, Sri Sairam Engineering College, West Tambaram, Chennai, India
| [b]
Senior Clinical Teaching Fellow in Obstetrics and Gynecology, South Warwickshire NHS Foundation Trust, Warwick Hospital, Warwick CV34 5BW, United Kingdom
Correspondence:
[*]
Corresponding author. R. Geetha, Research Scholar, Department of Electronics and Communication Engineering, Sri Sairam Engineering College, West Tambaram, Chennai, India. E-mail: [email protected].
Abstract: Purpose:
Automated diagnosis of acute cerebral ischemic stroke lesions (ACISL) is an evolving science. Early detection and exact delineation of ACISL automatically from diffusion-weighted magnetic resonance (DWMR) images are crucial for initiating prompt treatment. Thus, this work aims to determine the appropriate slice out of 60 pieces using multi-fractal analysis (MFA) and to segment the lesions in DWMR images using a hybrid optimization method. Features extracted from the segmented images were clinically correlated with the modified Rankin Scale (mRS). Methods:
Thirty-one real-time stroke patients’ images were collected from Rajiv Gandhi Government General Hospital, Chennai, India. Multiple MRI slices were taken from each patient and filtered using an anisotropic diffusion filter (ADF). These filtered images were skull-stripped automatically by the maximum entropy thresholding technique incorporating mathematical morphological operations (MEM). The multi-fractal analysis (MFA) identifies the prominent slice with the significant infarct lesion. An isodata algorithm that integrated differential evolution with the particle swarm optimization method based on Kapur’s (IDPK) and Otsu’s (IDPO) approaches was attempted to segment the ACISL. Finally, the geometric and moment features extracted from the segmented lesions categorized the stroke severity and were correlated with the mRS. Results:
The findings of the experimental work confirm that the suggested IDPK approach achieved usual normalized values for image similarity indices such as Sokal-Michener Coefficient (98.51%), Roger-Tanimoto Coefficient (90.16%), Sokel-Sneath-2 (91.04%), and Sorenson Index (90.04%) are superior to IDPO. Statistical significance proved that the segmented lesions’ area (r = 0.820, p < 0.0001) and perimeter (r = 0.928, p < 0.0001) were strongly correlated with the mild and moderate criteria of mRS. Conclusion:
The proposed work effectively detected ischemic stroke lesions and their severity within the studied image groups. It could be a promising and potential tool to aid radiologists in validating their diagnosis.
Keywords: Ischemic stroke lesion, magnetic resonance imaging, multi-fractal analysis, isodata algorithm, differential evolution with particle swarm optimization, modified Rankin Scale
