AI in Medicine
Homework #2
AI作业代写 In Homework #2 mini-project assignment, you will use synthetic middle sagittal MR images to build and train a convolutional…
Figure 1: Types of pelvic organ prolapse. a) Normal (no prolapse); b) AVP (anterior prolapse); c) PVP (posterior prolapse); d) UP (uterine prolapse)
Note: AI作业代写
- In Homework #2 mini-project assignment, you will use synthetic middle sagittal MR images to build and train a convolutional neural networks (classifier) for pelvic organ prolapse classification, with an average F1-score above 0.5 and F1-score of the normal group above 0.6.
- ‘synthetic_MR_classification_dataset’ is for this classification task
- ‘synthetic_MR_landmark_dataset’ is for the optional landmark localization task
- More details will be introduced in class.
Submission files (to course.pku.edu or email to both TA and instructor):
- Jupyter notebook with code or python file with code
- PDF of the code
- A writeup PDF report:
o The report would describe the main steps that you used to AI作业代写
- explore the dataset
- design and validate your classifier
- training and validation curves
- use your classifier to make predictions
o Please include the necessary code or figures
- (10 points) Data Exploration (using python code):
a.(6 points) Dataset Summary
i.Number of samples in each category
ii.Shape of the sample images ie. (x, x)
iii. Number of classes/labels
b.(4 points) Exploratory Visualization
i.For each class/label, plot a sample image
2.(90 points) Design and Test a Classifier (or model architecture) AI作业代写
a.(10 points) Preprocess the dataset using techniques such as normalization, resizing, and explain why you choose those techniques in report
b.(15 points) Design your model architecture with at least 5 layers, and show the architecture in your report; if you use one advanced CNN architecture for transfer learning, please explain why you choose that architecture
c.(10 points) Compile your model: choose the loss function, optimizer, and metrics, batch size, number of epochs, and other relevant values of hyperparameters
d.(10 points) Train your model, and plot your training history
e.(20 points) Tune your model (tune hyperparameters, or apply data augmentation), and save your final model, and plot the training and validation curves
f.(15 points) Evaluate your final model’s performance, and make predictions on 4 randomly selected test image
g.(5 points) Plot the ROC curve on the test dataset
h.(5 points) Plot the confusion matrix on the test dataset
3.(optional bonus: 20 points) AI作业代写
a.(10 points) Visualize selected the convolutional neural networks of the test images to help you understand what features your model extracted from the images (tips: you might use GradCAM)
b.(10 points) Change your output of the CNN model you just built, and apply to the ‘synthetic_MR_landmark_dataset’ for landmark-localization task. Please try to achieve an average MSE error of less than 50 pixels.
Figure 2: Two pelvic landmarks.
Appendix: (from Royal College of Obstetricians and Gynecologists)
Figure 3: Anatomical images and types of pelvic organ prolapse. a) Normal (no prolapse); b)
AVP (anterior prolapse); c) PVP (posterior prolapse); d) UP (uterine prolapse)
