Algorithm 1. Multi-Dataset YOLOv8 Training Pipeline

Input: ISIC 2020, HAM10000, PH2 Datasets

Output: Trained YOLOv8 Model for Melanoma Detection and Segmentation

1:Load Datasets:

2:  Import ISIC 2020, HAM10000, and PH2 datasets.

3:   Apply dataset balancing using stratified sampling.

4:   Normalize pixel intensity values to [0, 1] range.

5:  end

6:Preprocessing

7:  Resize all images to 512 × 512 resolution.

8:   Apply histogram matching to standardize color distribution.

9:   Remove artifacts using morphological operations.

10:  end

11:Data Augmentation

12:  Start Apply geometric transformations (rotation, flipping, scaling).

13:   Adjust brightness and contrast.

14:   Use advanced augmentation (CutMix, Mosaic) to improve generalization.

15:  end

16:Training (YOLOv8 Model)

17:  Initialize model with pre-trained weights.

18:   Use multi-dataset training with dataset weighting.

19:   Train using Adam optimizer and learning rate scheduling.

20:   Apply batch normalization and dropout to prevent overfitting.

21:  end

22:Post-Processing

23:  Start Apply segmentation boundary refinement.

24:   Filter false positives using confidence thresholding.

25:  end

26: Performance Evaluation

27:  Start Compute mAP@0.5, Dice Coefficient, and IoU.

28:   Compare results with U-Net, DeepLabV3+, and Mask R-CNN.

29:  end

30:Model Deployment

31:  Start Optimize for real-time inference.

32:   Test deployment on clinical and mobile health applications.

33:  end

34:End Algorithm