Cancer Detection from Microscopic-Tissue Images with Deep Learning (Auto ML, Custom Convolutional Neural Network, and Transfer Learning)

Domain             : Computer Vision, Machine Learning
Sub-Domain         : Deep Learning, Image Recognition
Techniques         : Deep Convolutional Neural Network, Transfer Learning, ImageNet, Auto ML, NASNetMobile
Application        : Image Recognition, Image Classification, Medical Imaging

Description

1. Detected Cancer from microscopic tissue images (histopathologic) with Auto ML (Google’s “NASNet”).
2. For training, concatenated global pooling (max, average), dropout and dense layers to the output layer for final output prediction.
3. Attained testing accuracy of 93.72% and loss 0.30 on 250K+ (6.5GB+) image cancer dataset.

Code

GitHub Link      : Histopathologic Cancer Detection(GitHub)
GitLab Link      : Histopathologic Cancer Detection(GitLab)
Portfolio        : Anjana Tiha's Portfolio

Dataset

Dataset Name     : Histopathologic Cancer Detection
Dataset Link     : Histopathologic Cancer Detection (Kaggle)
                 : PatchCamelyon (PCam) (GitHub)
                 : CAMELYON16 challenge Dataset (Original Dataset)
                 
Original Paper   : Diagnostic Assessment of Deep Learning Algorithms for Detection of Lymph Node Metastases in Women With Breast Cancer  
                   Authors: Babak Ehteshami Bejnordi, Mitko Veta, Paul Johannes van Diest 
                   JAMA (The Journal of the American Medical Association)
                   Ehteshami Bejnordi B, Veta M, Johannes van Diest P, et al. Diagnostic Assessment of Deep Learning Algorithms for Detection of Lymph Node Metastases in Women With Breast Cancer. JAMA. 2017;318(22):2199–2210. doi:10.1001/jama.2017.14585

Dataset Details

Dataset Name            : Histopathologic Cancer Detection
Number of Class         : 2

Dataset Subtype	Number of Image	Size of Images (GB/Gigabyte)
Total	220,025	5.72 GB
Training	132,016	3.43 GB
Validation	44,005	1.14 GB
Testing	44,004	1.14 GB

Model and Training Prameters

Current Parameters	Value
Base Model	NashNetLarge
Optimizers	Adam
Loss Function	Categorical Crossentropy
Learning Rate	0.0001
Batch Size	16
Number of Epochs	2
Training Time	4.5 hour (270 min)

Model Performance Metrics (Prediction/ Recognition / Classification)

Dataset	Training	Validation	Test
Accuracy	94.74%	93.62%	93.72%
Loss	0.14	0.30	0.30
Precision	---	---	89.02%
Recall	---	---	90.80%
Roc-Auc	---	---	91.59%

Other Experimented Model and Training Prameters

Parameters (Experimented)	Value
Base Models	NashNet(NashNetLarge, NashNetMobile), InceptionV3
Optimizers	Adam, SGD
Loss Function	Categorical Crossentropy, Binary Crossentropy
Learning Rate	0.0001, 0.00001, 0.000001, 0.0000001
Batch Size	16, 32, 64, 128, 256
Number of Epochs	2, 4, 6, 10, 30, 50, 100
Training Time	4.5 hour (270 min), 1 day (24 hours), 2 days (24 hours)

Sample Output:

See More Images

Confusion Matrix:

Tools / Libraries

Languages               : Python
Tools/IDE               : Anaconda
Libraries               : Keras, TensorFlow, Inception, ImageNet

Dates

Duration                : November 2018 - March 2019
Current Version         : v1.0.0.8
Last Update             : 03.24.2019