Image-Reconstruction Technique

Medical Image Reconstruction - Dermoscopic images

Overview

An official implementation of a Single image reconstruction technique for Melanoma Skin lesion images for feature extraction using PyTorch.

Requirments

Matlab R2019

Python 3.10.10

PyTorch 1.4

Pillow 5.1.0

scikit-image 0.19.3

numpy 1.14.5

This was tested on Python 3.7. To install the required packages, use the provided requirements.txt file like so:

pip install -r requirements.txt

Datasets

ISIC Challenge Datasets 2020 https://challenge.isic-archive.com/data/#2020

PH2 Dataset https://www.dropbox.com/s/k88qukc20ljnbuo/PH2Dataset.rar

Pre-trained model

MELLiResNet

MELIIGAN

https://drive.google.com/file/d/1FV0T8C_0Z6oMUuvOq9ELs6EsXqxLuS5O/view?usp=share_link

Input Size changing

The provided model was trained on ISIC 2019, PH2 dataset and mednode image inputs, but to run it on inputs of arbitrary size, you'll have to change the input shape as given.

from tensorflow import keras

# Load the model
model = keras.models.load_model('models/generator.h5')

# Define arbitrary spatial dims, and 3 channels.
inputs = keras.Input((None, None, 3))

# Trace out the graph using the input:
outputs = model(inputs)

# Override the model:
model = keras.models.Model(inputs, outputs)

# Now you are free to predict on images of any size.

Experimental Results

The experimental results on the benchmark datasets.

Quantitave Results

Algorithm		Bicubic	ESPCN	SRGAN	ESRGAN	MELIIGAN (MY model)
ISIC 2020	PSNR	22.42	23.61	25.03	27.28	32.87
Dataset	SSIM	0.7304	0.7602	0.7941	0.8203	0.8249
PH2	PSNR	22.02	23.21	25.73	28.88	32.89
Dataset	SSIM	0.7164	0.7462	0.7601	0.7863	0.9100
Med node	PSNR	21.23	23.5	24.20	28.63	33.49
Dataset	SSIM	0.6504	0.7002	0.7941	0.8003	0.9082

Qualitative Results

Comments

The queries and comments on my codes can be forwarded to [email protected]

Contributors

Made with contributors-img