Medical Image Classification
LIDC-IDRI
- The Lung Image Database Consortium (LIDC) and Image Database Resource Initiative (IDRI): a completed reference database of lung nodules on CT scans.
- Highly accurate model for prediction of lung nodule malignancy with CT scans
Bookmarks
- Review of CAD for lung cancer
- meet the following requirements:
- improve the performance of radiologists providing high sensitivity in the diagnosis
- a low number of false positives (FP)
- high processing speed
- present high level of automation
- low cost (of implementation, training, support and maintenance)
- the ability to detect different types and shapes of nodules
- software security assurance
- meet the following requirements:
Evaluation
https://machinelearningmastery.com/metrics-evaluate-machine-learning-algorithms-python/ https://www.ritchieng.com/machine-learning-evaluate-classification-model/
Training Model
Image Classification using Python and Scikit-learn Python machine learning: Introduction to image classification
Online Courses
- Stanford CS class CS231n: Convolutional Neural Networks for Visual Recognition
- MIT: Biomedical Signal and Image Processing
- https://blog.paralleldots.com/data-science/research-papers-image-classification/
SPIE-AAPM Lung CT Challenge
classification
quantitative image analysis methods for the diagnostic classification of malignant and benign lung nodules
- TCIA Dataset
- Readings:
- Google Cloud Citations
Collection Statistics | |
---|---|
Modalities | CT |
Number of Patients | 70 |
Number of Studies | 70 |
Number of Series | 70 |
Number of Images | 22,489 |
Images Size (GB) | 12.1 |
- training set (10 subjects) - CT-Training
- test set (60 subjects) - LUNGx
LUng Nodule Analysis 2016 (LUNA16)
detection
- Nodule detection (NDET) Using raw CT scans, the goal is to identify locations of possible nodules, and to assign a probability for being a nodule to each location. The pipeline typically consists of two stages: candidate detection and false positive reduction.
- False positive reduction (FPRED) Given a set of candidate locations, the goal is to assign a probability for being a nodule to each candidate location. Hence, one could see this as a classification task: nodule or not a nodule. Candidate locations will be provided in world coordinates. This set detects 1,162/1,186 nodules.
Data Science Bowl 2017
classification
determine when lesions in the lungs are cancerous
Lung CT Segmentation Challenge 2017 (LCTSC)
detection
comparison of various auto-segmentation algorithms
Collection Statistics | |
---|---|
Modalities | CT, RT |
Number of Patients | 60 |
Number of Studies | 60 |
Number of Series | 96 |
Number of Images | 9,569 |
Images Size (GB) | 4.8 |
LungCT-Diagnosis
detection, prediction
extract prognostic image features that will describe lung adenocarcinomas and will associate with overall survival
Collection Statistics | |
---|---|
Modalities | CT |
Number of Patients | 61 |
Number of Studies | 61 |
Number of Series | 61 |
Number of Images | 4682 |
Images Size (GB) | 2.5 |
Dataset
Data downloaded from cancerimagingarchive.net.
sudo apt-get install icedtea-netx
javaws /path/to/your.jnlp
Breast Cancer
https://www.kaggle.com/junkal/breast-cancer-prediction-using-machine-learning https://medium.com/@Petuum/deep-learning-for-breast-cancer-identification-from-histopathological-images-f38de0a658a5