Melanoma image classification - Neural Designer example
=======================================================

Contents
--------
melanoma_dataset_bmp/   900 ISIC 2016 dermoscopic images, 300 x 300 RGB BMP,
                        one folder per class (benign, malignant).
manifest.csv            Source member, label and SHA-256 of every image.
model/                  Trained Neural Designer 8.0.1 model:
                          melanoma.ndm (configuration, saved data split and
                          training settings), melanoma_melanoma_dataset_bmp_Params.bin
                          (trained parameters) and melanoma_predict.bat
                          (command-line inference generated by Neural Designer).
scores.csv              Malignant score calculated by Neural Designer 8.0.1 for
                        every image, with its class and subset
                        (training / validation / testing).
SOURCE.md               Provenance and preparation of the images.
LICENSES/               Dataset license notice (CC0 1.0).

Calculate the output for one image (Windows, Neural Designer 8.0.1 installed)
-----------------------------------------------------------------------------
    model\melanoma_predict.bat path\to\lesion.bmp --table

The script prints the benign and malignant scores. Its "Predicted class" uses
a threshold of 0.5. The example's operating point flags an image when the
malignant score is 0.11 or higher; that threshold was chosen on the validation
images. Prepare new images exactly as described in SOURCE.md (RGB,
300 x 300, Lanczos resampling, BMP).

Educational research example only. Not a medical device and not for diagnosis.
