Early Melonama Detetction
Some facts about skin cancer:
-
Every year there are more new cases of skin cancer than the combined incidence of cancers of the prostate, lung, breast and colon. Roughly 87,110 new cases of invasive melanoma has been diagnosed in the U.S. per year.
-
Early Melonama detection is critical!
- The estimated 5-year survival rate for patients whose melanoma is detected early is about 98 percent in the U.S.
- The survival rate falls to 62 percent when the disease reaches the lymph nodes, and
- The survival rate falls to 18 percent when the disease metastasizes to distant organs.
The recent emergence of machine learning and deep learning methods for medical image analysis has enabled the development of intelligent medical imaging-based diagnosis systems that can assist physicians in making better decisions about a patient’s health. In particular, skin imaging is a field where these new methods can be applied with a high rate of success.
Taxonomy/ Keywords : Medical Image Analysis, Deep Learning, Medical Decision Support Systems,
U-Net Convolutional Neural Networks, Transfer Learning,VGG-Net Convolutional Neural Networks, Melanoma, Dermoscopy, Skin Lesions, Skin Cancer, CoreML, TensorflowLite, MaskRCNN.
"Fine-tuned Smart Solution approach" (perfectly_segmented_classification) focuses on solving the problem of automatic skin lesion detection, particularly on melanoma detection, by " semantic segmentation and followed by classification (powered by Transfer Learning)" from dermoscopic images (from ISIC) using a deep learning based approach.
- For the 1st part of the solution approach (semantic segmentation for an accurate extraction of the lesion region) - ** U-Net convolutional neural network architecture** is applied.
- For the 2nd part of the solution approach (Classification /binary classification of "Benign versus Malignant Cancer" for early melanoma detection) -
- The proposed solution is built around the VGG-Net ConvNet architecture and
- uses the Transfer learning paradigm (to deal with the general challenge ofMinimal Ground Truth by avoid groud up training of Detection Model.).
P.S.
- This solution approach performs way better, optimized and offers better result. The experimental results of the segmentation evaluations demonstrate that the proposed method can outperform other state-of-the-art models.
- It very effectively deals with the general challenge of Minimal Ground Truth (i.e., challenge of lack of large amount of labelled dataset) by avoid groud up training of Detection Model.
- The model is general enough to be extended to multi-class skin lesion classification.
- "Classical Solution approach" (unaltered_classification) focuses on solving the problem of automatic skin lesion detection by straightway subjecting the entire image to classification alone model.
- "ROC Curve" is drawn as well for a comparative evaluation of classification alone (using the entire image) against a combination of the two approaches (segmentation followed by classification) in order to assess which of them achieves better classification results.
- Experimental results are encouraging: on the ISIC Archive dataset, the Smart Solution approach achieves a sensitivity value of 78.66%, which is significantly higher than the many of the state of the art (SOTA Models) on that dataset.
Overall Solution Approache Schemes
Dataset {ISIC : International Skin Imaging Collaboration}
The datasets (carrying Both malignant and benign lesions) to use are:
- ISIC_UDA-2_1
- ISIC_UDA-1_1
- ISIC_MSK-2_1
- ISIC_MSK-1_2
- ISIC_MSK-1_1
| Benign Images | Malignant Images |
|---|---|
| 1208 | 849 |
The Challneges The Challenge:
{ISBI: The International Symposium on Biomedical Imaging}
NOTE:
- ISBI 2016 challenge provides training data (~900 images) for participants to engage in all 3 components of lesion image analysis. A separate test dataset (~350 images) will be provided for participants to generate and submit automated results.
-
- ISBI 2018 - ISIC Archive 2018 -contains over 13,000 dermoscopic images.
| Participant | Rank | Accuracy | Dice | Jaccard | Sensitivity | Specificity |
|---|---|---|---|---|---|---|
| Our Model | (1st)* | 0.9176 | 0.8689 | 0.9176 | 0.9301 | 0.9544 |
| Urko Sanchez | 1st | 0.953 | 0.910 | 0.843 | 0.910 | 0.965 |
| Lequan Yu | 2nd | 0.949 | 0.897 | 0.829 | 0.911 | 0.957 |
(*) Metrics - Participants from the ISBI 2016 Challenge were ranked and awards granted based only on the Jaccard index.
| Method | Accuracy | Sensitivity | Precision |
|---|---|---|---|
| Unaltered lesion classification | 0.8469 | 0.8243 | 0.9523 |
| Perfectly segmented lesion classification | 0.8390 | 0.8648 | 0.9621 |
| Automatically segmented lesion class | 0.8174 | 0.8918 | 0.9681 |
-
Improve model accuracy
- with additional annotated data
- with Optimized Model implementation, such as Mask RCNN, or, Detectron, or, Detectron2, or, DETER {DETR Model - DEtection TRansformer- A Transformer based Object Detection Approach}.
- An Evolution journey of RCNN Family: Faster RCNN(Y’2015) ---> Mask RCNN (Y’2017) / YOLOV2 (Y’2017)-> Detctron (Y’2018) -> Detectron2(Y’2019) -> DETER model(Y’2020)
-
MLOPS (Enable Iterative Experimentation)
-
Create PWA using ReactJS
-
Publish IoS App (using CoreML ) & Android App (using TensorFlow Lite )
- Use Core ML to integrate machine learning models into IoS app. Core ML provides a unified representation for all models. Your app uses Core ML APIs and user data to make predictions, and to train or fine-tune models, all on the user’s device.
- Use TensorFlow Lite to deploy machine learning models on mobile and IoT devices and also for Quantization of the model.
-
Fast forward to recent times: The ISIC Archive contains over 23k images of skin lesions, labeled as 'benign' or 'malignant'. The archive can be found here: https://www.isic-archive.com/#!/onlyHeaderTop/gallery
-- https://www.isic-archive.com/#!/onlyHeaderTop/gallery
-- This script gives "Comfortable alternative script approach" to the current ways to download the archive, provided by the ISIC foundation and which are known to me, are the following:
- Download the entire archive via the direct download button on their website.
- Download all the partitions of the archive, called 'datasets' one by one
- Downloading the images one by one via the Grider API provided in the site
Generally options above are laborious or at times do not work, ISIC Archive Downloader script script provides a comfortable alternative.
- This script can download the entire ISIC archive (or parts of it/optional-download-abilities)
- All you have to do is run
python download_archive.py