Upload 2 files

Colony_Analyzer_AI_zstack2_HF.py

@@ -0,0 +1,356 @@
+#!/usr/bin/env python3
+# -*- coding: utf-8 -*-
+"""
+Created on Thu Mar 20 14:23:27 2025
+
+@author: mattc
+"""
+
+import os
+import cv2
+#this is the huggingface version
+def cut_img(img):
+    img_map = {}
+    width, height = img.size
+    i_num = height // 512
+    j_num = width // 512
+    count = 1
+    for i in range(i_num):
+        for j in range(j_num):
+            cropped_img = img.crop((512*j, 512*i, 512*(j+1), 512*(i+1)))
+            img_map[count] = cropped_img
+            #print(type(cropped_img))
+            count += 1
+    return img_map
+    
+import numpy as np
+
+def stitch(img_map):
+    rows = [
+        np.hstack([img_map[1], img_map[2], img_map[3], img_map[4]]),  # First row (images 0 to 3)
+        np.hstack([img_map[5], img_map[6], img_map[7], img_map[8]]),  # Second row (images 4 to 7)
+        np.hstack([img_map[9], img_map[10], img_map[11], img_map[12]])  # Third row (images 8 to 11)
+    ]
+    # Stack rows vertically
+    return(np.vstack(rows))
+
+
+from PIL import Image
+
+
+import matplotlib.pyplot as plt
+
+def visualize_segmentation(mask, image=0):
+    plt.figure(figsize=(10, 5))
+
+    if(not np.isscalar(image)):
+        # Show original image if it is entered
+        plt.subplot(1, 2, 1)
+        plt.imshow(image)
+        plt.title("Original Image")
+        plt.axis("off")
+
+    # Show segmentation mask
+    plt.subplot(1, 2, 2)
+    plt.imshow(mask, cmap="gray")  # Show as grayscale
+    plt.title("Segmentation Mask")
+    plt.axis("off")
+
+    plt.show()
+
+import torch
+from transformers import SegformerForSemanticSegmentation
+# Load fine-tuned model
+model = SegformerForSemanticSegmentation.from_pretrained("ReyaLabColumbia/Segformer_Colony_Counter")  # Adjust path
+device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
+model.to(device)
+model.eval()  # Set to evaluation mode
+
+# Load image processor
+from transformers import SegformerForSemanticSegmentation, SegformerImageProcessor
+image_processor = SegformerImageProcessor.from_pretrained("nvidia/segformer-b3-finetuned-cityscapes-1024-1024")
+
+def preprocess_image(image):
+    image = image.convert("RGB")  # Open and convert to RGB
+    inputs = image_processor(image, return_tensors="pt")  # Preprocess for model
+    return image, inputs["pixel_values"]
+
+def postprocess_mask(logits):
+    mask = torch.argmax(logits, dim=1)  # Take argmax across the class dimension
+    return mask.squeeze().cpu().numpy()  # Convert to NumPy array
+
+
+def eval_img(image):
+    # Load and preprocess image
+    image, pixel_values = preprocess_image(image)
+    pixel_values = pixel_values.to(device)
+    with torch.no_grad():  # No gradient calculation for inference
+        outputs = model(pixel_values=pixel_values)  # Run model
+        logits = outputs.logits
+    # Convert logits to segmentation mask
+    segmentation_mask = postprocess_mask(logits)
+    #visualize_segmentation(segmentation_mask,image)
+    segmentation_mask = cv2.resize(segmentation_mask, (512, 512), interpolation=cv2.INTER_LINEAR_EXACT)
+    return(segmentation_mask)
+
+def find_colonies(mask, size_cutoff, circ_cutoff):
+    binary_mask = np.where(mask == 1, 255, 0).astype(np.uint8)
+    contours, _ = cv2.findContours(binary_mask, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
+    contoursf = []
+    areas = []
+    for x in contours:
+        area = cv2.contourArea(x)
+        if (area < size_cutoff):
+            continue
+        perimeter = cv2.arcLength(x, True)
+
+        # Avoid division by zero
+        if perimeter == 0:
+            continue
+        
+        # Calculate circularity
+        circularity = (4 * np.pi * area) / (perimeter ** 2)
+        if circularity >= circ_cutoff:
+            contoursf.append(x)
+            areas.append(area)
+    return(contoursf, areas)
+
+def find_necrosis(mask):
+    binary_mask = np.where(mask == 2, 255, 0).astype(np.uint8)
+    contours, _ = cv2.findContours(binary_mask, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
+    return(contours)
+
+# contour_image = np.zeros_like(p)
+# contours =  find_necrosis(p)
+# cv2.drawContours(contour_image, contours, -1, (255), 2)
+# visualize_segmentation(contour_image)
+import pandas as pd
+def compute_centroid(contour):
+    M = cv2.moments(contour)
+    if M["m00"] == 0:  # Avoid division by zero
+        return None
+    cx = int(M["m10"] / M["m00"])
+    cy = int(M["m01"] / M["m00"])
+    return (cx, cy)
+
+
+def contours_overlap_using_mask(contour1, contour2, image_shape=(1536, 2048)):
+    """Check if two contours overlap using a bitwise AND mask."""
+    import numpy as np
+    import cv2
+    mask1 = np.zeros(image_shape, dtype=np.uint8)
+    mask2 = np.zeros(image_shape, dtype=np.uint8)
+
+
+    # Draw each contour as a white shape on its respective mask
+    cv2.drawContours(mask1, [contour1], -1, 255, thickness=cv2.FILLED)
+    cv2.drawContours(mask2, [contour2], -1, 255, thickness=cv2.FILLED)
+
+
+    # Compute bitwise AND to find overlapping regions
+    overlap = cv2.bitwise_and(mask1, mask2)
+    
+    return np.any(overlap)
+
+def analyze_colonies(mask, size_cutoff, circ_cutoff):
+    colonies,areas = find_colonies(mask, size_cutoff, circ_cutoff)
+    necrosis = find_necrosis(mask)
+    
+    data = []
+    
+    for x in range(len(colonies)):
+        colony = colonies[x]
+        colony_area = areas[x]
+        centroid = compute_centroid(colony)
+
+        # Check if any necrosis contour is inside the colony
+        necrosis_area = 0
+        nec_list =[]
+        for nec in necrosis:
+            # Check if the first point of the necrosis contour is inside the colony
+            if contours_overlap_using_mask(colony, nec):
+                nec_area = cv2.contourArea(nec)
+                necrosis_area += nec_area
+                nec_list.append(nec)
+
+        data.append({
+            "colony_area": colony_area,
+            "necrosis_area": necrosis_area,
+            "centroid": centroid,
+            "percent_necrosis": necrosis_area/colony_area,
+            "contour": colony,
+            "nec_contours": nec_list
+        })
+
+    # Convert results to a DataFrame
+    df = pd.DataFrame(data)
+    df.index = range(1,len(df.index)+1)
+    return(df)
+
+
+def contour_overlap(contour1, contour2, centroid1, centroid2, area1, area2, centroid_thresh=30, area_thresh = .4, img_shape = (1536, 2048)):
+    """
+    Determines the overlap between two contours.
+    Returns:
+        0: No overlap
+        1: Overlap but does not meet strict conditions
+        2: Overlap >= 80% of the larger contour and centroids are close
+    """
+    # Create blank images
+    img1 = np.zeros(img_shape, dtype=np.uint8)
+    img2 = np.zeros(img_shape, dtype=np.uint8)
+    
+    # Draw filled contours
+    cv2.drawContours(img1, [contour1], -1, 255, thickness=cv2.FILLED)
+    cv2.drawContours(img2, [contour2], -1, 255, thickness=cv2.FILLED)
+    
+    # Compute overlap
+    intersection = cv2.bitwise_and(img1, img2)
+    intersection_area = np.count_nonzero(intersection)
+    
+    if intersection_area == 0:
+        return 0  # No overlap
+    
+    # Compute centroid distance
+    centroid_distance = float(np.sqrt(abs(centroid1[0]-centroid2[0])**2 + abs(centroid1[1]-centroid2[1])**2))
+    # Check percentage overlap relative to the larger contour
+    overlap_ratio = intersection_area/max(area1, area2)
+    if overlap_ratio >= area_thresh and centroid_distance <= centroid_thresh:
+        if area1 > area2:
+            return(2)
+        else:
+            return(3)
+    else:
+        return 1  # Some overlap but not meeting strict criteria
+    
+def compare_frames(frame1, frame2):
+    for i in range(1, len(frame1)+1):
+        if frame1.loc[i,"exclude"] == True:
+            continue
+        for j in range(1, len(frame2)+1):
+            if frame2.loc[j,"exclude"] == True:
+                continue
+            temp = contour_overlap(frame1.loc[i, "contour"], frame2.loc[j, "contour"], frame1.loc[i, "centroid"], frame2.loc[j, "centroid"], frame1.loc[i, "colony_area"], frame2.loc[j, "colony_area"])
+            if temp ==2:
+                frame2.loc[j,"exclude"] = True
+            elif temp ==3:
+                frame1.loc[i, "exclude"] = True
+                break
+    frame1 = frame1[frame1["exclude"]==False]
+    frame2 = frame2[frame2["exclude"]==False]
+    df = pd.concat([frame1, frame2], axis=0)
+    df.index = range(1,len(df.index)+1) 
+    return(df)
+    
+def main(args):
+    min_size = args[1]
+    min_circ = args[2]
+    colonies = {}
+    files = args[0]
+    for idx,x in enumerate(files):
+        img_map = cut_img(files[idx])
+        for z in img_map:
+            img_map[z] = eval_img(img_map[z])
+        del z
+        p = stitch(img_map)
+        frame = analyze_colonies(p, min_size, min_circ)
+        frame["source"] = idx
+        frame["exclude"] = False
+        if isinstance(colonies, dict):
+            colonies = frame
+        else:
+           colonies = compare_frames(frame, colonies)
+    counts = {}
+    for x in range(len(files)):
+        counts[x] = list(colonies["source"]).count(x)
+    best = [x, counts[x]]
+    del x
+    for x in counts:
+        if counts[x] > best[1]:
+            best[0] = x
+            best[1] = counts[x]
+    del x, counts
+    best = best[0]
+    img = np.array(files[best])
+    for x in range(len(files)):
+        if x == best:
+            continue
+        mask = np.zeros_like(cv2.cvtColor(img, cv2.COLOR_BGR2GRAY))
+        contours = colonies[colonies["source"]==x]
+        contours = list(contours["contour"])
+        cv2.drawContours(mask, contours, -1, 255, thickness=cv2.FILLED)
+        # Extract all ROIs from the source image at once
+        src_image = np.array(files[x])
+        roi = cv2.bitwise_and(src_image, src_image, mask=mask)
+        # Paste the extracted regions onto the destination image
+        np.copyto(img, roi, where=(mask[..., None] == 255))
+    try:
+        del x, mask, src_image, roi, best, contours
+    except:
+        pass
+    
+    img = cv2.copyMakeBorder(img,top=0, bottom=10,left=0,right=10, borderType=cv2.BORDER_CONSTANT,  value=[255, 255, 255]) 
+    colonies = colonies.sort_values(by=["colony_area"], ascending=False)
+    colonies = colonies[colonies["colony_area"]>= min_size]
+    colonies.index = range(1,len(colonies.index)+1) 
+    #nearby is a boolean list of whether a colony has overlapping colonies. If so, labelling positions change
+    nearby = [False]*len(colonies)
+    areas = list(colonies["colony_area"])
+    for i in range(len(colonies)): 
+        cv2.drawContours(img, [list(colonies["contour"])[i]], -1, (0, 255, 0), 2)
+        cv2.drawContours(img, list(colonies['nec_contours'])[i], -1, (0, 0, 255), 2)
+        coords = list(list(colonies["centroid"])[i])
+        if coords[0] > 1950:
+            #if a colony is too close to the right edge, makes the label move to left
+            coords[0] = 1950
+        for j in range(len(colonies)):
+            if j == i:
+                continue
+            coords2 = list(list(colonies["centroid"])[j])
+            if ((abs(coords[0] - coords2[0]) + abs(coords[1] - coords2[1])) <=  40):
+                nearby[i] = True
+                break
+        if nearby[i] ==True:
+            #If the colony has nearby colonies, this adjusts the labels so they are smaller and are positioned based on the approximate radius of the colony
+            # a random number is generated, and based on that, the label is put at the top or bottom, left or right
+            radius= int(np.sqrt(areas[i]/3.1415)*.9)
+            n = np.random.random()
+            if n >.75:
+                new_x = min(coords[0] + radius, 2000)
+                new_y = min(coords[1] + radius, 1480)
+            elif n >.5:
+                new_x = min(coords[0] + radius, 2000)
+                new_y = max(coords[1] - radius, 50)
+            elif n >.25:
+                new_x = max(coords[0] - radius, 0)
+                new_y = min(coords[1] + radius, 1480)
+            else:
+                new_x = max(coords[0] - radius, 0)
+                new_y = max(coords[1] - radius, 50)
+            cv2.putText(img, str(colonies.index[i]), (new_x,new_y), cv2.FONT_HERSHEY_SIMPLEX, 1, (0, 0, 0), 2)
+            del n, radius, new_x, new_y
+        else:
+            cv2.putText(img, str(colonies.index[i]), coords, cv2.FONT_HERSHEY_SIMPLEX, 2, (0, 0, 0), 2)
+    del nearby, areas
+    colonies = colonies.drop('contour', axis=1)
+    colonies = colonies.drop('nec_contours', axis=1)
+    colonies = colonies.drop('exclude', axis=1)
+    img = cv2.copyMakeBorder(img,top=10, bottom=0,left=10,right=0, borderType=cv2.BORDER_CONSTANT,  value=[255, 255, 255]) 
+    
+    colonies.insert(loc=0, column="Colony Number", value=[str(x) for x in range(1, len(colonies)+1)])
+    total_area_dark = sum(colonies['necrosis_area'])
+    total_area_light = sum(colonies['colony_area'])
+    ratio = total_area_dark/(abs(total_area_light)+1)
+
+    colonies.loc[len(colonies)+1] = ["Total", total_area_light, total_area_dark, None, ratio, None]
+    Parameters = pd.DataFrame({"Minimum colony size in pixels":[min_size], "Minimum colony circularity":[min_circ]})
+    with pd.ExcelWriter("Group_analysis_results.xlsx") as writer:
+        colonies.to_excel(writer, sheet_name="Colony data", index=False)
+        Parameters.to_excel(writer, sheet_name="Parameters", index=False)
+    caption = np.ones((150, 2068, 3), dtype=np.uint8) * 255  # Multiply by 255 to make it white
+    cv2.putText(caption, "Total area necrotic: "+str(total_area_dark)+ ", Total area living: "+str(total_area_light)+", Ratio: "+str(ratio), (40, 40), cv2.FONT_HERSHEY_SIMPLEX, 1, (0, 0, 0), 3)
+
+
+
+    cv2.imwrite('Group_analysis_results.png', np.vstack((img, caption)))
+    return(np.vstack((img, caption)), 'Group_analysis_results.png', 'Group_analysis_results.xlsx')

app.py

@@ -1,33 +1,56 @@
 import gradio as gr
-import Colony_Analyzer_AI2_HF as analyzer
 from PIL import Image
-import cv2
-import numpy as np
 
-# Analysis function adapted from your Tkinter app
+# Single image analysis function (your existing logic)
 def analyze_image(image, min_size, circularity):
-    # Assume your analyzer.main accepts [image, params] format, adjust as needed
-    processed_img,picname, excelname = analyzer.main([image, min_size, circularity])
-    print(type(processed_img))
-    # Convert back to RGB for display
-    #result = cv2.cvtColor(processed_img, cv2.COLOR_BGR2RGB)
+    import Colony_Analyzer_AI2_HF as analyzer
+    processed_img, picname, excelname = analyzer.main([image, min_size, circularity])
     return Image.fromarray(processed_img), picname, excelname
 
-# Create Gradio interface
-iface = gr.Interface(
-    fn=analyze_image,
-    inputs=[
-        gr.Image(type="pil", label="Upload Image"),
-        gr.Number(label="Minimum Colony Size (pixels)", value=1000),
-        gr.Number(label="Minimum Circularity", value=0.25)
-    ],
-    outputs=[
-        gr.Image(type="pil", label="Analyzed Image"),
-        gr.File(label="Download Image"),
-        gr.File(label="Download results (Excel)")
-    ],
-    title="AI Colony Analyzer",
-    description="Upload an image to run the colony analysis."
-)
-
-iface.launch()
+# Z-stack analysis function (adapt with your own logic)
+def analyze_zstack(images, min_size, circularity):
+    # images: list of PIL images
+    # Plug in your own z-stack segmentation logic here
+    # Example stub: pass images as a list to your analyzer
+    import Colony_Analyzer_AI_zstack2_HF as analyzer
+    processed_img, picname, excelname = analyzer.main([images, min_size, circularity])
+    return Image.fromarray(processed_img), picname, excelname
+
+with gr.Blocks() as demo:
+    gr.Markdown("# AI Colony Analyzer\nUpload an image (or Z-Stack) to run colony analysis.")
+
+    z_stack_checkbox = gr.Checkbox(label="Enable Z-Stack", value=False)
+    image_input_single = gr.Image(type="pil", label="Upload Image", visible=True)
+    image_input_multi = gr.Image(type="pil", label="Upload Z-Stack Images", file_count="multiple", visible=False)
+    min_size_input = gr.Number(label="Minimum Colony Size (pixels)", value=1000)
+    circularity_input = gr.Number(label="Minimum Circularity", value=0.25)
+    output_image = gr.Image(type="pil", label="Analyzed Image")
+    output_file_img = gr.File(label="Download Image")
+    output_file_excel = gr.File(label="Download results (Excel)")
+    process_btn = gr.Button("Process")
+
+    def toggle_inputs(z_stack_enabled):
+        return (
+            gr.update(visible=not z_stack_enabled),  # single input
+            gr.update(visible=z_stack_enabled)        # multi input
+        )
+
+    z_stack_checkbox.change(
+        toggle_inputs,
+        inputs=z_stack_checkbox,
+        outputs=[image_input_single, image_input_multi]
+    )
+
+    def conditional_analyze(z_stack, single_image, multi_images, min_size, circularity):
+        if z_stack:
+            return analyze_zstack(multi_images, min_size, circularity)
+        else:
+            return analyze_image(single_image, min_size, circularity)
+
+    process_btn.click(
+        conditional_analyze,
+        inputs=[z_stack_checkbox, image_input_single, image_input_multi, min_size_input, circularity_input],
+        outputs=[output_image, output_file_img, output_file_excel]
+    )
+
+demo.launch()