Biomedical images are increasing drastically. Along the way, many machine
learning algorithms have been proposed to predict and identify various kinds of
diseases. One such disease is Pneumonia which is an infection caused by both
bacteria and viruses through the inflammation of a person's lung air sacs. In
this paper, an algorithm was proposed that receives x-ray images as input and
verifies whether this patient is infected by Pneumonia as well as specific
region of the lungs that the inflammation has occurred at. The algorithm is
based on the transfer learning mechanism where pre-trained ResNet-50
(Convolutional Neural Network) was used followed by some custom layer for
making the prediction. The model has achieved an accuracy of 90.6 percent which
confirms that the model is effective and can be implemented for the detection
of Pneumonia in patients. Furthermore, a class activation map is used for the
detection of the infected region in the lungs. Also, PneuNet was developed so
that users can access more easily and use the services.
Predicting Pneumonia and Region Detection from X-Ray Images using
X線画像からの肺炎の予測と領域検出
0.83
Deep Neural Network
ディープニューラルネットワーク
0.68
Sheikh Md Hanif Hossain1, S M Raju2, Amelia Ritahani Ismail3
Sheikh Md Hanif Hossain1, S M Raju2, Amelia Ritahani Ismail3
0.96
Department of Computer Science International Islamic University Malaysia
計算機科学専攻 国際イスラム大学マレーシア
0.66
1sheikhhanifhossain@ gmail.com, 2rajuiium121@gmail.c om, 3amelia@iium.edu.my
1sheikhhanifhossain@ gmail.com, 2rajuiium121@gmail.c om, 3amelia@iium.edu.my
0.59
– as In are well
– として 院 は まあ
0.64
sacs. kinds images Abstract
袋だ 種類 画像 概要
0.62
lung air Biomedical
肺空気 バイオメディカル
0.71
disease by such caused
病気が原因で その原因は
0.51
the way, many machine
ところで 多くの機械が
0.81
One infection the increasing drastically.
感染が1つ 劇的に増えています
0.63
Along learning algorithms have been proposed to predict and identify is diseases.
is疾患を予測し識別するために、学習アルゴリズムが提案されている。
0.61
various of is Pneumonia which an both bacteria and viruses inflammation of a through person’s this paper, an algorithm was proposed that receives x-ray images as input and verifies whether this patient is infected by Pneumonia as the inflammation has occurred at.
The algorithm is based learning on pre- transfer trained ResNet-50 (Convolutional Neural Network) was used followed by some custom layer for making the prediction.
The model has achieved an accuracy of is effective and can be implemented for the detection of Pneumonia class activation map is used for the detection of the infected region in the lungs.
Among some of to detect the pneumonia generating heatmap to see the regions where the pneumonia was detected transfer learning to detect whether the person has pneumonia or not two articles that transfer learning yielded better results than other used focal loss to prevent bias in the imbalanced dataset [3].
Many used Class Activation Map (CAM) to get the discriminative image regions used by identify a a Convolutional Neural Network (CNN) specific a trained on Convolutional Neural Network which was millions thousand classify used classes of images accurately [5].
Data augmentation was performed to increase the diversity of the training data without inserting new data into the training image flipping, and rotation were done to increase the number of training records.
web application where the user will upload the image of x- ray and the model will identify and predict whether the uploaded x-ray image has pneumonia or not.
And if the patient has pneumonia then the system will identify the region patient highlighted the infected region in his x-ray image.
患者が肺炎を患っている場合、システムはx線画像で感染した領域を強調した領域の患者を識別する。
0.74
deployed another image
配備 もう一つ image
0.58
show and the
ショー そして はあ?
0.59
as to a IV.
として へ あ IV。
0.61
Results The final model has achieved an accuracy of 90.06 percent using the test dataset for predicting whether a patient has been infected by pneumonia.
Furthermore, the precision and recall of our model are 92 and 93
さらに、モデルの精度とリコールは92と93である。
0.73
3.3 Model final architecture.
3.3モデル 最後の建築
0.64
The custom followed by a dropout the output
慣習に従って出力がドロップアウトされる
0.72
In this paper, pre-trained ResNet-50 was used as a base model.
本稿では,ResNet-50をベースモデルとして使用した。
0.73
Image size of 224,224,3 was used as input for base model.
224,224,3の画像サイズをベースモデルの入力として用いた。
0.65
The base model layers were frozen and followed by some custom layers were added to create the layers consist of a layer and again a dense dense layer as well.
The first dense layer had 50 neurons and for the activation function Rectified Linear Unit (ReLU) was used.
最初の高密度層は50個のニューロンを持ち、活性化関数Rectified Linear Unit (ReLU) が使用された。
0.78
Then in the dropout layer, dropout 0.5 was applied to all the neurons.
そして、ドロップアウト層において、すべてのニューロンに0.5のドロップアウトを施した。
0.59
In the final layer, sigmoid activation functions were used to get the final output of the model.
最終層では、モデルの最終出力を得るためにsgmoidアクティベーション関数が使用された。
0.78
The layer layer which was
layer (複数形 layers)
0.47
Figure 1: Train vs validation loss respectively.
図1: 電車とバリデーションの損失それぞれ。
0.68
From Figure 1, it percent, seen how the train and validation loss is decreasing over the number of is increasing over the number of epochs for both training and in
Characteristic (ROC) curve was shown and the Area Under Curve (AUC) was 89 percent.
特性曲線(ROC)が示され,AUC(Area Under Curve)は99%であった。
0.76
Figure 3: Confusion matrix
図3:混乱マトリックス
0.72
Figure 4: ROC Curve Figure 5: PneuNet prediction
図4:ROC曲線 図5:PneuNet予測
0.79
Figure 5 shows that the patient who uploaded the x- images is
図5は、X-画像をアップロードした患者が
0.78
pneumonia ray the highlighted using the PneuNet system.
pneumonia ray the highlight using the pneunet system (英語)
0.77
infected area and
感染 エリア そして
0.73
has V. Discussion
have ... V.討論
0.51
percent. percent 92
パーセント 92パーセントです
0.53
90 previous 77 around that
90年前 77人です
0.59
This is researches [1].
これがresearches [1]である。
0.69
and From the result section, the accuracy of this model around huge is showing a over their improvement as Also, around was accuracy the precision recall are and percent, 93 respectively.