import streamlit as st import tensorflow as tf from PIL import Image import numpy as np import cv2 import matplotlib.pyplot as plt from imutils import perspective from scipy.spatial import distance as dist model=tf.keras.models.load_model("dental_xray_seg.h5") st.header("Segmentation of Teeth in Panoramic X-ray Image") examples=["teeth_01.png","teeth_02.png","teeth_03.png","teeth_04.png"] def load_image(image_file): img = Image.open(image_file) img_gray = img.convert('L') img_np = np.array(img_gray) return img_np def convert_one_channel(img): if len(img.shape)>2: img= cv2.cvtColor(img, cv2.COLOR_BGR2GRAY) return img def convert_rgb(img): if len(img.shape)==2: img= cv2.cvtColor(img,cv2.COLOR_GRAY2RGB) return img def midpoint(ptA, ptB): return ((ptA[0] + ptB[0]) * 0.5, (ptA[1] + ptB[1]) * 0.5) def CCA_Analysis(orig_image,predict_image,erode_iteration,open_iteration): kernel1 =( np.ones((5,5), dtype=np.float32)) kernel_sharpening = np.array([[-1,-1,-1], [-1,9,-1], [-1,-1,-1]]) image = predict_image image2 =orig_image image=cv2.morphologyEx(image, cv2.MORPH_OPEN, kernel1,iterations=open_iteration ) image = cv2.filter2D(image, -1, kernel_sharpening) image=cv2.erode(image,kernel1,iterations =erode_iteration) image=cv2.cvtColor(image, cv2.COLOR_BGR2GRAY) thresh = cv2.threshold(image, 0, 255, cv2.THRESH_BINARY + cv2.THRESH_OTSU)[1] labels=cv2.connectedComponents(thresh,connectivity=8)[1] a=np.unique(labels) count2=0 for label in a: if label == 0: continue # Create a mask mask = np.zeros(thresh.shape, dtype="uint8") mask[labels == label] = 255 # Find contours and determine contour area cnts,hieararch = cv2.findContours(mask.copy(), cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE) cnts = cnts[0] c_area = cv2.contourArea(cnts) # threshhold for tooth count if c_area>2000: count2+=1 (x,y),radius = cv2.minEnclosingCircle(cnts) rect = cv2.minAreaRect(cnts) box = cv2.boxPoints(rect) box = np.array(box, dtype="int") box = perspective.order_points(box) color1 = (list(np.random.choice(range(150), size=3))) color =[int(color1[0]), int(color1[1]), int(color1[2])] cv2.drawContours(image2,[box.astype("int")],0,color,2) (tl,tr,br,bl)=box (tltrX,tltrY)=midpoint(tl,tr) (blbrX,blbrY)=midpoint(bl,br) # compute the midpoint between the top-left and top-right points, # followed by the midpoint between the top-righ and bottom-right (tlblX,tlblY)=midpoint(tl,bl) (trbrX,trbrY)=midpoint(tr,br) # draw the midpoints on the image cv2.circle(image2, (int(tltrX), int(tltrY)), 5, (255, 0, 0), -1) cv2.circle(image2, (int(blbrX), int(blbrY)), 5, (255, 0, 0), -1) cv2.circle(image2, (int(tlblX), int(tlblY)), 5, (255, 0, 0), -1) cv2.circle(image2, (int(trbrX), int(trbrY)), 5, (255, 0, 0), -1) cv2.line(image2, (int(tltrX), int(tltrY)), (int(blbrX), int(blbrY)),color, 2) cv2.line(image2, (int(tlblX), int(tlblY)), (int(trbrX), int(trbrY)),color, 2) dA = dist.euclidean((tltrX, tltrY), (blbrX, blbrY)) dB = dist.euclidean((tlblX, tlblY), (trbrX, trbrY)) pixelsPerMetric=1 dimA = dA * pixelsPerMetric dimB = dB *pixelsPerMetric cv2.putText(image2, "{:.1f}pixel".format(dimA),(int(tltrX - 15), int(tltrY - 10)), cv2.FONT_HERSHEY_SIMPLEX,0.65, color, 2) cv2.putText(image2, "{:.1f}pixel".format(dimB),(int(trbrX + 10), int(trbrY)), cv2.FONT_HERSHEY_SIMPLEX,0.65, color, 2) cv2.putText(image2, "{:.1f}".format(label),(int(tltrX - 35), int(tltrY - 5)), cv2.FONT_HERSHEY_SIMPLEX,0.65, color, 2) teeth_count=count2 return image2,teeth_count def detect_decays_static_th(images, dental_masks=None, threshhold=0.9): decay_masks = [] for image, dental_mask in zip(images, dental_masks): decay_mask = np.zeros_like(dental_mask) image_masked_with_dental_mask = image * dental_mask decay_mask[image_masked_with_dental_mask > threshhold*255] = 1 decay_masks.append(decay_mask) decay_masks = np.array(decay_masks) return decay_masks st.subheader("Upload Dental Panoramic X-ray Image") image_file = st.file_uploader("Upload Images", type=["png","jpg","jpeg"]) col1, col2, col3, col4 = st.columns(4) with col1: ex=load_image(examples[0]) st.image(ex,width=200) if st.button('Example 1'): image_file=examples[0] with col2: ex1=load_image(examples[1]) st.image(ex1,width=200) if st.button('Example 2'): image_file=examples[1] with col3: ex2=load_image(examples[2]) st.image(ex2,width=200) if st.button('Example 3'): image_file=examples[2] with col4: ex2=load_image(examples[3]) st.image(ex2,width=200) if st.button('Example 4'): image_file=examples[3] if image_file is not None: image_original = Image.open(image_file) image=np.asarray(image_original) image = convert_rgb(image) st.subheader("Original Image") st.image(image,width=1100) st.text("Making A Prediction ....") img=np.asarray(image) img_cv=convert_one_channel(img) img_cv=cv2.resize(img_cv,(512,512), interpolation=cv2.INTER_LANCZOS4) img_cv=np.float32(img_cv/255) img_cv=np.reshape(img_cv,(1,512,512,1)) prediction=model.predict(img_cv) predicted=prediction[0] predicted_rgb = np.expand_dims(predicted, axis=-1) plt.imsave("predict.png",predicted_rgb) predict1 = cv2.resize(predicted, (img.shape[1], img.shape[0]), interpolation=cv2.INTER_LANCZOS4) img_dc=convert_one_channel(img) decay_mask = detect_decays_static_th(img_dc, predict1) mask = np.uint8(predict1 * 255) _, mask = cv2.threshold(mask, thresh=255/2, maxval=255, type=cv2.THRESH_BINARY) cnts, hierarchy = cv2.findContours(mask, cv2.RETR_TREE, cv2.CHAIN_APPROX_SIMPLE) img = cv2.drawContours(img, cnts, -1, (0, 0, 255), 2) mask = np.uint8(decay_mask * 255) _, mask = cv2.threshold(mask, thresh=255/2, maxval=255, type=cv2.THRESH_BINARY) cnts, hierarchy = cv2.findContours(mask, cv2.RETR_TREE, cv2.CHAIN_APPROX_SIMPLE) img = cv2.fillPoly(img, cnts, (255, 0, 0)) if img is not None : st.subheader("Predicted teeth shape + caries zones") st.write(img.shape) st.image(img,width=1100) image=np.asarray(image_original) image = convert_rgb(image) if image.shape[1] < 3000: image = cv2.resize(image,(3100,1150),interpolation=cv2.INTER_LANCZOS4) predicted=cv2.imread("predict.png") predicted = cv2.resize(predicted, (image.shape[1],image.shape[0]), interpolation=cv2.INTER_LANCZOS4) cca_result,teeth_count=CCA_Analysis(image,predicted,3,2) if cca_result is not None : st.subheader("Seperate predicted teeth") st.write(cca_result.shape) st.image(cca_result,width=1100) st.text("Teeth Count = " + str(teeth_count)) st.text("DONE ! ....")