Merge pull request #291 from AIM-Harvard/v1

Update docs

README.md

@@ -1,15 +1,22 @@
 
 <center>
-This is the the official repository for the paper: "<b>Foundation Model for Cancer Imaging Biomarkers </b>"
+This is the the official repository for the paper
+
+<br/>
+
 </center>
-<div style="display: flex; justify-content: center"><img src="docs/assets/Header.png"/>
+<div style="display: flex; flex-direction: column; align-items: center;">
+    <img src="./docs/assets/Header.png" style="width: 100%;"/>
+    <div style="display: flex; justify-content: space-between; width: 100%;">
+        <img src="./docs/assets/Mhub_image.png" style="width: 50%;"/>
+        <img src="./docs/assets/Mhub_image2.png" style="width: 50%;"/>
+    </div>
+    <a href="https://www.nature.com/articles/s42256-024-00807-9"><img src="./docs/assets/readpaper_logo.png" style="width: 100%;"></a>
 </div>
+
 <br/><br/>
 <div align="center">
 
-<i><font size="-1">Suraj Pai, Dennis Bontempi, Ibrahim Hadzic, Vasco Prudente, Mateo Sokač, Tafadzwa L. Chaunzwa, Simon Bernatz, Ahmed Hosny, Raymond H Mak, Nicolai J Birkbak, Hugo JWL Aerts</i></font>
-
-
 [![Build Status](https://github.com/AIM-Harvard/foundation-cancer-image-biomarker/actions/workflows/build.yml/badge.svg)](https://github.com/AIM-Harvard/foundation-cancer-image-biomarker/actions/workflows/build.yml)
 [![Python Version](https://img.shields.io/pypi/pyversions/foundation-cancer-image-biomarker.svg)](https://pypi.org/project/foundation-cancer-image-biomarker/)
 [![Dependencies Status](https://img.shields.io/badge/dependencies-up%20to%20date-brightgreen.svg)](https://github.com/AIM-Harvard/foundation-cancer-image-biomarker/pulls?utf8=%E2%9C%93&q=is%3Apr%20author%3Aapp%2Fdependabot)
@@ -23,7 +30,7 @@ This is the the official repository for the paper: "<b>Foundation Model for Canc
 </div>
 
 ---
-**NOTE: **
+**NOTE:**
  For detailed documentation check our [website](https://aim-harvard.github.io/foundation-cancer-image-biomarker/) 
 
 ---

docs/index.md

@@ -3,14 +3,13 @@ hide:
   - title
 ---
 # 
-
-<center>
-This is the the official documentation for the paper: "<b>Foundation Model for Cancer Imaging Biomarkers </b>"
-</center>
-<div style="display: flex;"><img src="assets/Header.png"/></div>
-
-
-<i><font size="-1">Suraj Pai, Dennis Bontempi, Ibrahim Hadzic, Vasco Prudente, Mateo Sokač, Tafadzwa L. Chaunzwa, Simon Bernatz, Ahmed Hosny, Raymond H Mak, Nicolai J Birkbak, Hugo JWL Aerts</i></font>
+<div style="display: flex; flex-direction: column; align-items: center;">
+    <img src="assets/Header.png" style="width: 100%;"/>
+    <div style="display: flex; justify-content: space-between; width: 100%;">
+        <img src="assets/Mhub_image.png" style="width: 50%;"/>
+        <img src="assets/Mhub_image2.png" style="width: 50%;"/>
+    </div>
+</div>
 
 
 ## Documentation Walkthrough
@@ -20,11 +19,11 @@ This is the the official documentation for the paper: "<b>Foundation Model for C
 !!! note
     [We also provide quickstart examples that run in a free-cloud based environment](./getting-started/cloud-quick-start.md) (through Google Colab) so you can get familiar with our workflows, without having to download anything on your local machine!!
 
-[Replication Guide](./user-guide/data.md)<br> If you would like to pre-train a foundation model on your own unannotated data or would like to replicate the training and evaluation from our study, see here. 
+[Replication Guide](./replication-guide/data.md)<br> If you would like to pre-train a foundation model on your own unannotated data or would like to replicate the training and evaluation from our study, see here. 
 
 [Tutorials](https://github.com/AIM-Harvard/foundation-cancer-image-biomarker/tree/master/tutorials)<br> We provide comprehensive tutorials that use the foundation model for cancer imaging biomarkers and compare against other popularly used methods. If you would like to build your own study using our foundation model, these set of tutorials are highly recommended as the starting point. 
 
-[API Docs](./api_docs/fmcib/index.html) <br> This is for the more advanced user who would like to deep-dive into different methods and classes provided by our package. 
+[API Docs](./reference/run) <br> This is for the more advanced user who would like to deep-dive into different methods and classes provided by our package. 
 
 
 ## License

docs/replication-guide/baselines.md

@@ -0,0 +1,2 @@
+# Reproduce Baselines
+:hourglass_flowing_sand: Coming soon! :hourglass_flowing_sand:

docs/user-guide/data.md → docs/replication-guide/data.md

@@ -62,7 +62,7 @@ bash luna16.sh <path_to_download>
 The easiest way to download the LUNG1 and RADIO datasets is through s5cmd and [IDC manifests](https://learn.canceridc.dev/data/downloading-data)
 For convenience, the manifests for each of the already been provided in `data/download` under `nsclc_radiomics.csv` for LUNG1 and `nsclc_radiogenomics.`
 
-First, you'll need to install `s5cmd`. Follow the instructions here: https://github.com/peak/s5cmd?tab=readme-ov-file#installation
+First, you'll need to install `s5cmd`. Follow the instructions [here]https://github.com/peak/s5cmd?tab=readme-ov-file#installation
 
 Once you have s5cmd installed, run 
 

fmcib/visualization/verify_io.py

@@ -17,45 +17,54 @@ def visualize_seed_point(row):
         None
     """
     # Define the transformation pipeline
+    is_label_provided = "label_path" in row
+    keys = ["image_path", "label_path"] if is_label_provided else ["image_path"]
+    all_keys = keys if is_label_provided else ["image_path", "coordX", "coordY", "coordZ"]
+
     T = monai_transforms.Compose(
         [
-            monai_transforms.LoadImaged(keys=["image_path"], image_only=True, reader="ITKReader"),
-            monai_transforms.EnsureChannelFirstd(keys=["image_path"]),
-            monai_transforms.Spacingd(keys=["image_path"], pixdim=1, mode="bilinear", align_corners=True, diagonal=True),
+            monai_transforms.LoadImaged(keys=keys, image_only=True, reader="ITKReader"),
+            monai_transforms.EnsureChannelFirstd(keys=keys),
+            monai_transforms.Spacingd(keys=keys, pixdim=1, mode="bilinear", align_corners=True, diagonal=True),
             monai_transforms.ScaleIntensityRanged(keys=["image_path"], a_min=-1024, a_max=3072, b_min=0, b_max=1, clip=True),
-            monai_transforms.Orientationd(keys=["image_path"], axcodes="LPS"),
-            monai_transforms.SelectItemsd(keys=["image_path", "coordX", "coordY", "coordZ"]),
+            monai_transforms.Orientationd(keys=keys, axcodes="LPS"),
+            monai_transforms.SelectItemsd(keys=all_keys),
         ]
     )
 
     # Apply the transformation pipeline
     out = T(row)
 
     # Calculate the center of the image
-    center = (-out["coordX"], -out["coordY"], out["coordZ"])
-    center = np.linalg.inv(np.array(out["image_path"].affine)) @ np.array(center + (1,))
-    center = [int(x) for x in center[:3]]
-
-    # Define the image and label
     image = out["image_path"]
-    label = torch.zeros_like(image)
-
-    # Define the dimensions of the image and the patch
-    C, H, W, D = image.shape
-    Ph, Pw, Pd = 50, 50, 50
-
-    # Calculate and clamp the ranges for cropping
-    min_h, max_h = max(center[0] - Ph // 2, 0), min(center[0] + Ph // 2, H)
-    min_w, max_w = max(center[1] - Pw // 2, 0), min(center[1] + Pw // 2, W)
-    min_d, max_d = max(center[2] - Pd // 2, 0), min(center[2] + Pd // 2, D)
-
-    # Check if coordinates are valid
-    assert min_h < max_h, "Invalid coordinates: min_h >= max_h"
-    assert min_w < max_w, "Invalid coordinates: min_w >= max_w"
-    assert min_d < max_d, "Invalid coordinates: min_d >= max_d"
-
-    # Define the label for the cropped region
-    label[:, min_h:max_h, min_w:max_w, min_d:max_d] = 1
+    if not is_label_provided:
+        center = (-out["coordX"], -out["coordY"], out["coordZ"])
+        center = np.linalg.inv(np.array(out["image_path"].affine)) @ np.array(center + (1,))
+        center = [int(x) for x in center[:3]]
+
+        # Define the image and label
+        label = torch.zeros_like(image)
+
+        # Define the dimensions of the image and the patch
+        C, H, W, D = image.shape
+        Ph, Pw, Pd = 50, 50, 50
+
+        # Calculate and clamp the ranges for cropping
+        min_h, max_h = max(center[0] - Ph // 2, 0), min(center[0] + Ph // 2, H)
+        min_w, max_w = max(center[1] - Pw // 2, 0), min(center[1] + Pw // 2, W)
+        min_d, max_d = max(center[2] - Pd // 2, 0), min(center[2] + Pd // 2, D)
+
+        # Check if coordinates are valid
+        assert min_h < max_h, "Invalid coordinates: min_h >= max_h"
+        assert min_w < max_w, "Invalid coordinates: min_w >= max_w"
+        assert min_d < max_d, "Invalid coordinates: min_d >= max_d"
+
+        # Define the label for the cropped region
+        label[:, min_h:max_h, min_w:max_w, min_d:max_d] = 1
+    else:
+        label = out["label_path"]
+        center = torch.nonzero(label).float().mean(dim=0)
+        center = [int(x) for x in center][1:]
 
     # Blend the image and the label
     ret = blend_images(image=image, label=label, alpha=0.3, cmap="hsv", rescale_arrays=False)

mkdocs.yml

@@ -39,7 +39,8 @@ plugins:
           docstring_style: google
           options:
             # Removed the default filter that excludes private members (that is, members whose names start with a single underscore).
-            filters: null      
+            filters: null     
+            show_source: true 
 
 nav:
   - 'index.md'
@@ -48,12 +49,13 @@ nav:
     - 'Cloud Quick Start': 'getting-started/cloud-quick-start.md'
     - 'Quick Start': 'getting-started/quick-start.md'
   - 'Replication Guide':
-    - 'Data Download and Preprocessing': 'user-guide/data.md'
-    - 'Pre-training the FM': 'user-guide/reproduce_fm.md'
-    - 'Adapt the FM to downstream tasks': 'user-guide/fm_adaptation.md'
-    - 'Extracting Features & Predictions': 'user-guide/inference.md'
-    - 'Reproduce Analysis': 'user-guide/analysis.md'
-    # - 'Training baselines': 'user-guide/reproduce_baselines.md'
+    - 'Data Download and Preprocessing': 'replication-guide/data.md'
+    - 'Pre-training the FM': 'replication-guide/reproduce_fm.md'
+    - 'Adapt the FM to downstream tasks': 'replication-guide/fm_adaptation.md'
+    - 'Baselines for downstream tasks': 'replication-guide/baselines.md'
+    - 'Extracting Features & Predictions': 'replication-guide/inference.md'
+    - 'Reproduce Analysis': 'replication-guide/analysis.md'
+    # - 'Training baselines': 'replication-guide/reproduce_baselines.md'
   - 'Tutorials': https://github.com/AIM-Harvard/foundation-cancer-image-biomarker/tree/master/tutorials
   - 'API Reference': 'reference/'
 

scripts/generate_api_reference_pages.py

@@ -18,7 +18,10 @@
 root = Path(__file__).parent.parent
 src = root / PACKAGE
 
-for path in sorted(src.rglob("*.py")):
+# Sort files by depth
+paths = sorted(src.rglob("*.py"), key=lambda path: len(path.parts))
+
+for path in paths:
     print(f"Processing {path}")
     module_path = path.relative_to(src).with_suffix("")