Technical field of inyention;
(0002( The invention is to address the task of accurately segmenting skin lesions from medical
images using advanced deep learning technique. Specifically, it aims to automatically and
precisely delineate the boundaries of skin lesions, such as melanoma, moles, and other
dermatological abnormalities, in medical images.
Summary of the invention;
(0003( Data Pre-Processing- A comprehensive dataset of skin lesion images is gathered and preprocessed
to maintain data integrity and uniformity. Normalization and scaling of the data are
conducted to remove biases and support efficient optimization.
(0004( Meta heuristics Optimisation - The proposed technique uses state-of-the-art metaheuristics
like Improved Random Parameter-based Galactic Swarm Optimization (IRP-GSO) to intelligently
explore the feature space and find the most important input features. The metaheuristics
dynamically adapt to the problem at hand, providing efficient and accurate feature selection.
(0005( Skin Lesion Segmentation - The gated attention mechanism in GAUNET++ allows the
model to focus on relevant and salient features within the skin lesion, enabling more precise and
accurate segmentation. This can overcome the limitation of traditional segmentation methods that
may not effectively capture fine details or emphasize crucial regions. The invention also employs GAUNET++ that is frequently used to improve the accuracy of skin lesion segmentation by
adjusting parameters such as epochs, optimizer, steps per epoch, and activation function using
IRP-GSO.
Brief description of drawing;
There are three stages in this proposed work, which are explained in more detail below
a. An overview of data set
b. Pre-processing of data
c. Skin Lesion Segmentation
FIG 1 illustrates a system for the proposed architecture of segmentation model using Gated
Attention Unet++.
[0006] Enhanced Accuracy: By leveraging advanced optimisation methods, the technique
significantly improves the accuracy of skin lesion segmentation.
[0007] Intelligent Feature Selection: The adaptive nature of metaheuristics ensures that the most
influential input features are selected, reducing computational complexity and enhancing model
performance.
[0008) Advanced Segmentation Model: The invention drives skin lesion detection to new heights,
promoting efficient utilisation of skin lesion images.
IlWiic!aim;
I. The system claims I, wherein that claims selecting the best segmentation model for skin lesion
prediction system