Skip to content

Latest commit

 

History

History
98 lines (83 loc) · 5.29 KB

README.md

File metadata and controls

98 lines (83 loc) · 5.29 KB
Raw

Malaria Detection from Microscopic-Tissue Images with Deep Learning (NasNetMobile)

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

Description

1. Detected Malaria from microscopic tissue images by completely retraining pretrained model (Google's "NASNet") from scratch.
2. For training, concatenated global max pooling, global average pooling, flattened output, then added a dense layer with batch normalization and dropout and give to the output layer for final output prediction.
3. Attained validation accuracy of 95.72% and loss 0.1385 on 27K+ (330MB+) image malaria dataset.

Code

GitHub Link      : Malaria Detection using Deep Learning (GitHub)
GitLab Link      : Malaria Detection using Deep Learning (GitLab)
Kaggle Link      : Malaria Detection using Keras (Accuracy - 95%)
Portfolio        : Anjana Tiha's Portfolio

Dataset

Dataset Name     : Malaria Cell Images Dataset
Dataset Link     : Malaria Cell Images Dataset (Kaggle)
Original Dataset : Malaria Datasets -National Institutes of Health (NIH)

Dataset Details

Dataset Name            : Malaria Cell Images Dataset
Number of Class         : 2
Dataset Subtype Number of Image Size(MB/Megabyte)
Total 27,588 337 MB
Training 20,670 ---
Validation 6,888 ---
Testing --- ---
Dataset Subtype Number of Image Size of Images (GB/Gigabyte)
Total 27,588 337 MB
Training 20,670 ---
Validation 6,888 ---
Testing --- ---

Model and Training Prameters

Current Parameters Value
Base Model NashNetMobile
Optimizers Adam
Loss Function Categorical Crossentropy
Learning Rate 0.0001
Batch Size 176
Number of Epochs 10
Training Time 45 Min

Model Performance Metrics (Prediction/ Recognition / Classification)

Dataset Training Validation Test
Accuracy 96.47% 95.72% ---
Loss 0.1025 0.1385 ---
Precision --- --- ---
Recall --- --- ---
Roc-Auc --- --- ---

Other Experimented Model and Training Prameters

Parameters (Experimented) Value
Base Models NashNet(NashNetMobile)
Optimizers Adam
Loss Function Categorical Crossentropy
Learning Rate 0.0001, 0.00001, 0.000001, 0.0000001
Batch Size 32, 64, 176
Number of Epochs 10
Training Time 45 Min

Tools / Libraries

Languages               : Python
Tools/IDE               : Kaggle
Libraries               : Keras, TensorFlow, NasNetMobile

Dates

Duration                : February 2019 - Current
Current Version         : v1.0.0.9
Last Update             : 03.14.2019