move folder to this repo

README.md

@@ -1 +1,42 @@
-# covid-severity
+## Predicting COVID-19 Pneumonia Severity on Chest X-ray with Deep Learning
+
+🛑 NOT FOR MEDICAL USE 🛑 
+
+Data is here: [Pneumonia severity scores for 94 images](https://github.com/ieee8023/covid-chestxray-dataset/blob/master/annotations/covid-severity-scores.csv)
+
+  - **geographic_extent_mean**: The extent of lung involvement by ground glass opacity or consolidation for each lung. The total extent score ranged from 0 to 8.		
+  - **opacity_mean**: The degree of opacity. The total opacity score ranged from 0 to 6.
+
+License: CC BY-SA Creative Commons Attribution-ShareAlike
+
+These are from the follow paper:
+Cohen, Joseph Paul, et al. Predicting COVID-19 Pneumonia Severity on Chest X-Ray with Deep Learning. May 2020, http://arxiv.org/abs/2005.11856.
+
+```
+@article{Cohen2020Severity,
+title = {Predicting COVID-19 Pneumonia Severity on Chest X-ray with Deep Learning},
+author = {Cohen, Joseph Paul and Dao, Lan and Morrison, Paul and Roth, Karsten and Bengio, Yoshua and Shen, Beiyi and Abbasi, Almas and Hoshmand-Kochi, Mahsa and Ghassemi, Marzyeh and Li, Haifang and Duong, Tim Q},
+url = {http://arxiv.org/abs/2005.11856},
+year = {2020}
+}
+```
+
+To run the CLI:
+
+```
+$ python predict_severity.py 2966893D-5DDF-4B68-9E2B-4979D5956C8E.jpeg
+geographic_extent (0-8): 5.978744940174467
+opacity (0-6): 4.169582852893416
+
+# or to output a saliency map:
+$ python predict_severity.py 01E392EE-69F9-4E33-BFCE-E5C968654078.jpeg  -saliency_path heatmap.jpg
+
+```
+
+| Image             | Saliency map | Predictions       |
+|-------------------------------|-------------------------------|-----------|
+| ![](examples/2966893D-5DDF-4B68-9E2B-4979D5956C8E.jpeg-resize.jpg)| ![](examples/2966893D-5DDF-4B68-9E2B-4979D5956C8E.jpeg-heatmap.jpg)| geographic_extent (0-8): 5.979 <br>opacity (0-6): 4.17 |
+| ![](examples/31BA3780-2323-493F-8AED-62081B9C383B.jpeg-resize.jpg)| ![](examples/31BA3780-2323-493F-8AED-62081B9C383B.jpeg-heatmap.jpg)| geographic_extent (0-8): 6.293 <br>opacity (0-6): 4.367 |
+| ![](examples/41591_2020_819_Fig1_HTML.webp-day5.png-resize.jpg)| ![](examples/41591_2020_819_Fig1_HTML.webp-day5.png-heatmap.jpg)| geographic_extent (0-8): 3.067 <br>opacity (0-6): 2.335 |
+| ![](examples/8FDE8DBA-CFBD-4B4C-B1A4-6F36A93B7E87.jpeg-resize.jpg)| ![](examples/8FDE8DBA-CFBD-4B4C-B1A4-6F36A93B7E87.jpeg-heatmap.jpg)| geographic_extent (0-8): 0.9483 <br>opacity (0-6): 1.0 |
+

covid-severity-scores.csv

@@ -0,0 +1,100 @@
+## COVID-19 Pneumonia Severity Scores (Mean of 3 expert raters),,
+## License: CC BY-SA Creative Commons Attribution-ShareAlike,,
+## Citation: Joseph Paul Cohen et al. Predicting COVID-19 Pneumonia Severity on Chest X-Ray with Deep Learning. May 2020 http://arxiv.org/abs/2005.11856.,,
+## geographic_extent_mean: The extent of lung involvement by ground glass opacity or consolidation for each lung. The total extent score ranged from 0 to 8.,,
+## opacity_mean: The degree of opacity. The total opacity score ranged from 0 to 6.,,
+filename,geographic_mean,opacity_mean
+01E392EE-69F9-4E33-BFCE-E5C968654078.jpeg,6.0,4.0
+03BF7561-A9BA-4C3C-B8A0-D3E585F73F3C.jpeg,2.7,2.0
+1-s2.0-S0140673620303706-fx1_lrg.jpg,2.0,2.0
+1-s2.0-S1684118220300608-main.pdf-001.jpg,3.7,3.0
+1-s2.0-S1684118220300608-main.pdf-002.jpg,1.7,1.7
+1-s2.0-S1684118220300682-main.pdf-002-a1.png,0.0,0.0
+1-s2.0-S1684118220300682-main.pdf-002-a2.png,0.3,0.3
+1-s2.0-S1684118220300682-main.pdf-003-b1.png,1.3,1.3
+1-s2.0-S1684118220300682-main.pdf-003-b2.png,2.0,2.0
+1.CXRCTThoraximagesofCOVID-19fromSingapore.pdf-000-fig1a.png,1.0,1.0
+1.CXRCTThoraximagesofCOVID-19fromSingapore.pdf-000-fig1b.png,1.0,1.0
+1.CXRCTThoraximagesofCOVID-19fromSingapore.pdf-001-fig2a.png,0.3,0.3
+1.CXRCTThoraximagesofCOVID-19fromSingapore.pdf-001-fig2b.png,1.0,1.0
+1.CXRCTThoraximagesofCOVID-19fromSingapore.pdf-002-fig3a.png,1.0,1.0
+1.CXRCTThoraximagesofCOVID-19fromSingapore.pdf-002-fig3b.png,4.7,2.7
+1.CXRCTThoraximagesofCOVID-19fromSingapore.pdf-003-fig4a.png,4.0,3.3
+1.CXRCTThoraximagesofCOVID-19fromSingapore.pdf-003-fig4b.png,4.7,4.0
+1312A392-67A3-4EBF-9319-810CF6DA5EF6.jpeg,4.0,3.0
+1B734A89-A1BF-49A8-A1D3-66FAFA4FAC5D.jpeg,2.0,2.3
+23E99E2E-447C-46E5-8EB2-D35D12473C39.png,2.0,2.0
+2966893D-5DDF-4B68-9E2B-4979D5956C8E.jpeg,6.7,4.0
+2C26F453-AF3B-4517-BB9E-802CF2179543.jpeg,2.3,2.0
+31BA3780-2323-493F-8AED-62081B9C383B.jpeg,7.7,6.0
+353889E0-A1E8-4F9E-A0B8-F24F36BCFBFB.jpeg,4.7,4.0
+41591_2020_819_Fig1_HTML.webp-day10.png,2.3,2.7
+41591_2020_819_Fig1_HTML.webp-day5.png,2.3,3.0
+53EC07C9-5CC6-4BE4-9B6F-D7B0D72AAA7E.jpeg,5.0,3.3
+58cb9263f16e94305c730685358e4e_jumbo.jpeg,1.0,1.0
+6b44464d-73a7-4cf3-bbb6-ffe7168300e3.annot.original.jpeg,0.0,0.0
+6C94A287-C059-46A0-8600-AFB95F4727B7.jpeg,8.0,5.0
+6CB4EFC6-68FA-4CD5-940C-BEFA8DAFE9A7.jpeg,6.3,3.3
+7AF6C1AF-D249-4BD2-8C26-449304105D03.jpeg,2.0,2.7
+7C69C012-7479-493F-8722-ABC29C60A2DD.jpeg,0.0,0.0
+80446565-E090-4187-A031-9D3CEAA586C8.jpeg,3.7,2.0
+85E52EB3-56E9-4D67-82DA-DEA247C82886.jpeg,4.7,4.0
+8FDE8DBA-CFBD-4B4C-B1A4-6F36A93B7E87.jpeg,0.3,0.3
+93FE0BB1-022D-4F24-9727-987A07975FFB.jpeg,3.3,3.7
+9C34AF49-E589-44D5-92D3-168B3B04E4A6.jpeg,1.7,1.7
+9fdd3c3032296fd04d2cad5d9070d4_jumbo.jpeg,1.0,1.0
+all14238-fig-0001-m-b.jpg,6.0,4.0
+all14238-fig-0001-m-c.jpg,6.0,4.0
+auntminnie-a-2020_01_28_23_51_6665_2020_01_28_Vietnam_coronavirus.jpeg,2.3,1.3
+auntminnie-b-2020_01_28_23_51_6665_2020_01_28_Vietnam_coronavirus.jpeg,4.0,3.0
+auntminnie-c-2020_01_28_23_51_6665_2020_01_28_Vietnam_coronavirus.jpeg,4.7,3.0
+auntminnie-d-2020_01_28_23_51_6665_2020_01_28_Vietnam_coronavirus.jpeg,5.7,3.7
+B2D20576-00B7-4519-A415-72DE29C90C34.jpeg,6.7,3.3
+CD50BA96-6982-4C80-AE7B-5F67ACDBFA56.jpeg,3.0,2.7
+ciaa199.pdf-001-a.png,0.0,0.0
+ciaa199.pdf-001-b.png,0.0,0.0
+ciaa199.pdf-001-c.png,0.0,0.0
+covid-19-infection-exclusive-gastrointestinal-symptoms-pa.png,2.3,2.0
+covid-19-pneumonia-12.jpg,8.0,4.0
+covid-19-pneumonia-14-PA.png,3.3,2.0
+covid-19-pneumonia-15-PA.jpg,4.7,4.0
+covid-19-pneumonia-19.jpg,5.0,2.0
+covid-19-pneumonia-2.jpg,1.3,2.0
+covid-19-pneumonia-28.png,5.7,4.0
+covid-19-pneumonia-30-PA.jpg,3.0,2.0
+covid-19-pneumonia-7-PA.jpg,2.3,2.3
+covid-19-pneumonia-evolution-over-a-week-1-day0-PA.jpg,0.7,0.7
+covid-19-pneumonia-rapidly-progressive-12-hours.jpg,7.3,3.0
+covid-19-pneumonia-rapidly-progressive-3-days.jpg,8.0,3.0
+covid-19-pneumonia-rapidly-progressive-admission.jpg,4.3,3.0
+covid-19-rapidly-progressive-acute-respiratory-distress-syndrome-ards-admission.jpg,6.3,5.0
+covid-19-rapidly-progressive-acute-respiratory-distress-syndrome-ards-day-1.jpg,7.0,4.0
+covid-19-rapidly-progressive-acute-respiratory-distress-syndrome-ards-day-2.jpg,8.0,6.0
+E63574A7-4188-4C8D-8D17-9D67A18A1AFA.jpeg,3.3,2.0
+F2DE909F-E19C-4900-92F5-8F435B031AC6.jpeg,7.0,5.0
+F4341CE7-73C9-45C6-99C8-8567A5484B63.jpeg,4.0,2.0
+F63AB6CE-1968-4154-A70F-913AF154F53D.jpeg,5.7,4.0
+FE9F9A5D-2830-46F9-851B-1FF4534959BE.jpeg,3.3,2.0
+figure1-5e71be566aa8714a04de3386-98-left.jpeg,7.3,6.0
+figure1-5e75d0940b71e1b702629659-98-right.jpeg,0.0,0.0
+gr1_lrg-a.jpg,4.3,3.7
+gr1_lrg-b.jpg,2.0,1.3
+kjr-21-e24-g001-l-a.jpg,4.0,3.3
+kjr-21-e24-g002-l-a.jpg,3.0,2.3
+kjr-21-e24-g003-l-a.jpg,1.0,1.0
+kjr-21-e25-g001-l-a.jpg,0.7,0.7
+lancet-case2a.jpg,6.0,4.3
+lancet-case2b.jpg,7.3,5.0
+nCoV-radiol.2020200269.fig1-day7.jpeg,0.7,1.0
+nejmc2001573_f1a.jpeg,2.0,2.0
+nejmc2001573_f1b.jpeg,2.7,2.0
+nejmoa2001191_f1-PA.jpeg,0.0,0.0
+nejmoa2001191_f3-PA.jpeg,0.7,0.7
+nejmoa2001191_f4.jpeg,1.3,1.0
+nejmoa2001191_f5-PA.jpeg,2.0,2.3
+radiol.2020200490.fig3.jpeg,4.3,3.3
+ryct.2020200028.fig1a.jpeg,2.0,1.7
+ryct.2020200034.fig2.jpeg,1.3,1.3
+ryct.2020200034.fig5-day0.jpeg,3.7,3.3
+ryct.2020200034.fig5-day4.jpeg,5.7,4.0
+ryct.2020200034.fig5-day7.jpeg,4.3,4.0

predict_severity.py

@@ -0,0 +1,107 @@
+#!/usr/bin/env python
+# coding: utf-8
+
+import os,sys
+sys.path.insert(0,"..")
+from glob import glob
+import matplotlib.pyplot as plt
+import numpy as np
+import argparse
+import skimage
+import pprint
+
+import torch
+import torch.nn.functional as F
+import torchvision, torchvision.transforms
+import skimage, skimage.filters
+import torchxrayvision as xrv
+
+parser = argparse.ArgumentParser()
+parser.add_argument('img_path', type=str)
+parser.add_argument('-cuda', default=False, help='', action='store_true')
+parser.add_argument('-saliency_path', default=None, help='path to write the saliancy map as an image')
+
+cfg = parser.parse_args()
+
+
+img = skimage.io.imread(cfg.img_path)
+img = xrv.datasets.normalize(img, 255)  
+
+# Check that images are 2D arrays
+if len(img.shape) > 2:
+    img = img[:, :, 0]
+if len(img.shape) < 2:
+    print("error, dimension lower than 2 for image")
+
+# Add color channel
+img = img[None, :, :]                    
+
+
+transform = torchvision.transforms.Compose([xrv.datasets.XRayCenterCrop(),
+                                            xrv.datasets.XRayResizer(224)])
+
+img = transform(img)
+
+
+class PneumoniaSeverityNet(torch.nn.Module):
+    def __init__(self):
+        super(PneumoniaSeverityNet, self).__init__()
+        self.model = xrv.models.DenseNet(weights="all")
+        self.model.op_threshs = None
+        self.theta_bias_geographic_extent = torch.from_numpy(np.asarray((0.8705248236656189, 3.4137437)))
+        self.theta_bias_opacity = torch.from_numpy(np.asarray((0.5484423041343689, 2.5535977)))
+
+    def forward(self, x):
+        preds = self.model(x)
+        preds = preds[0,xrv.datasets.default_pathologies.index("Lung Opacity")]
+        geographic_extent = preds*self.theta_bias_geographic_extent[0]+self.theta_bias_geographic_extent[1]
+        opacity = preds*self.theta_bias_opacity[0]+self.theta_bias_opacity[1]
+        geographic_extent = torch.clamp(geographic_extent,0,8)
+        opacity = torch.clamp(opacity,0,6)
+        return {"geographic_extent":geographic_extent,"opacity":opacity}
+
+
+model2 = PneumoniaSeverityNet()
+
+with torch.no_grad():
+    img = torch.from_numpy(img).unsqueeze(0)
+    if cfg.cuda:
+        img = img.cuda()
+        model2 = model2.cuda()
+
+    outputs = model2(img)
+    print("geographic_extent (0-8):","{:1.4}".format(outputs["geographic_extent"].cpu().numpy()))
+    print("opacity (0-6):","{:1.4}".format(outputs["opacity"].cpu().numpy()))
+
+
+
+def full_frame(width=None, height=None):
+    import matplotlib as mpl
+    mpl.rcParams['savefig.pad_inches'] = 0
+    figsize = None if width is None else (width, height)
+    fig = plt.figure(figsize=figsize)
+    ax = plt.axes([0,0,1,1], frameon=False)
+    ax.get_xaxis().set_visible(False)
+    ax.get_yaxis().set_visible(False)
+    plt.autoscale(tight=True)
+
+if cfg.saliency_path:  
+    if cfg.cuda:
+        img = img.cuda()
+        model2 = model2.cuda()
+
+    img = img.requires_grad_()
+    outputs = model2(img)
+    grads = torch.autograd.grad(outputs["geographic_extent"], img)[0][0][0]
+    blurred = skimage.filters.gaussian(grads**2, sigma=(5, 5), truncate=3.5)
+
+    full_frame()
+    my_dpi = 100
+    fig = plt.figure(frameon=False, figsize=(224/my_dpi, 224/my_dpi), dpi=my_dpi)
+    ax = plt.Axes(fig, [0., 0., 1., 1.])
+    ax.set_axis_off()
+    fig.add_axes(ax)
+    ax.imshow(img[0][0].detach(), cmap="gray", aspect='auto')
+    ax.imshow(blurred, alpha=0.5);
+    plt.savefig(cfg.saliency_path)
+

requirements.txt

@@ -0,0 +1,5 @@
+numpy
+pandas
+torch
+torchvision
+torchxrayvision