【修复】新增对DICOM图像的支持 (#310)

* 修复对DICOM图像加载逻辑，允许从DICOM文件读取并应用窗口设置

* 新增DICOM的测试图片

* 统一了处理入口，标准化像素值，确保三通道输出

* 修复了单通道png图无法正常显示和显示的通道图与实际图片不一致的问题

Author: ShowUNow

ISAT/annotation.py

@@ -6,9 +6,56 @@
 import numpy as np
 from json import load, dump
 from typing import List, Union
+import pydicom
 
 __all__ = ['Object', 'Annotation']
 
+
+def get_windowed_image(ds):
+    # Apply Rescale Slope and Rescale Intercept if they exist
+    pixel_array = ds.pixel_array.astype(float)
+    rescale_slope = ds.get("RescaleSlope", 1)
+    rescale_intercept = ds.get("RescaleIntercept", 0)
+    if rescale_slope != 1 or rescale_intercept != 0:
+        pixel_array = pixel_array * rescale_slope + rescale_intercept
+
+    # Check for windowing information in the DICOM metadata
+    window_center = ds.get("WindowCenter", None)
+    window_width = ds.get("WindowWidth", None)
+
+    if window_center is None or window_width is None:
+        # If no windowing info, use the full dynamic range
+        window_min = pixel_array.min()
+        window_max = pixel_array.max()
+    else:
+        # Handle possible multi-valued tags by taking the first value
+        if isinstance(window_center, pydicom.multival.MultiValue):
+            window_center = window_center[0]
+        if isinstance(window_width, pydicom.multival.MultiValue):
+            window_width = window_width[0]
+
+        window_min = window_center - window_width / 2
+        window_max = window_center + window_width / 2
+
+    # Apply windowing
+    pixel_array = np.clip(pixel_array, window_min, window_max)
+
+    # Normalize to 0-255
+    if window_max > window_min:
+        pixel_array = ((pixel_array - window_min) / (window_max - window_min)) * 255.0
+    else:  # Handle case where all pixels are the same
+        pixel_array.fill(128)
+
+    # Handle Photometric Interpretation
+    photometric_interpretation = ds.get("PhotometricInterpretation", "MONOCHROME2")
+    if photometric_interpretation == "MONOCHROME1":
+        pixel_array = 255.0 - pixel_array
+
+    # Convert to 8-bit unsigned integer
+    image_8bit = pixel_array.astype(np.uint8)
+
+    return image_8bit
+
 class Object:
     r"""A class to represent an annotation object.
 
@@ -57,20 +104,48 @@ def __init__(self, image_path:str, label_path:str):
         self.img_name = img_name
         self.label_path = label_path
         self.note = ''
+        self.image_path = image_path
+        self._img_data = None
 
-        image = np.array(Image.open(image_path))
-        if image.ndim == 3:
-            self.height, self.width, self.depth = image.shape
-        elif image.ndim == 2:
-            self.height, self.width = image.shape
-            self.depth = 0
-        else:
-            self.height, self.width, self.depth = image.shape[:, :3]
-            print('Warning: Except image has 2 or 3 ndim, but get {}.'.format(image.ndim))
-        del image
+        # Defer image loading to get_img_data()
+        self.width = 0
+        self.height = 0
+        self.depth = 0
 
         self.objects:List[Object, ] = []
 
+    def get_img_data(self, to_rgb=False):
+        if self._img_data is None:
+            if self.image_path.lower().endswith('.dcm'):
+                ds = pydicom.dcmread(self.image_path)
+                image_data = get_windowed_image(ds)
+                self._img_data = np.stack([image_data, image_data, image_data], axis=-1)
+            else:
+                self._img_data = Image.open(self.image_path)
+
+        if self.width == 0 or self.height == 0:
+            if isinstance(self._img_data, np.ndarray):
+                image = self._img_data
+                if image.ndim == 3:
+                    self.height, self.width, self.depth = image.shape
+                elif image.ndim == 2:
+                    self.height, self.width = image.shape
+                    self.depth = 1
+            else: # PIL Image
+                self.width, self.height = self._img_data.size
+                self.depth = len(self._img_data.getbands())
+
+        if to_rgb:
+            if isinstance(self._img_data, np.ndarray):
+                return self._img_data # DICOM data is already RGB numpy array
+            else: # PIL Image
+                return np.array(self._img_data.convert('RGB'))
+        else:
+            if isinstance(self._img_data, np.ndarray):
+                 return self._img_data
+            else: # PIL Image
+                return np.array(self._img_data)
+
     def load_annotation(self):
         r"""
         Load annotation from self.label_path

ISAT/widgets/canvas.py

@@ -90,85 +90,21 @@ def load_image(self, image_path: str):
         if self.mainwindow.use_segment_anything:
             self.mainwindow.segany.reset_image()
 
-        if image_path.lower().endswith('.dcm'):
-            ds = pydicom.dcmread(image_path)
-            # Apply Rescale Slope and Rescale Intercept if they exist
-            pixel_array = ds.pixel_array.astype(float)
-            rescale_slope = ds.get("RescaleSlope", 1)
-            rescale_intercept = ds.get("RescaleIntercept", 0)
-            if rescale_slope != 1 or rescale_intercept != 0:
-                pixel_array = pixel_array * rescale_slope + rescale_intercept
-
-            # Check for windowing information in the DICOM metadata
-            window_center = ds.get("WindowCenter", None)
-            window_width = ds.get("WindowWidth", None)
-
-            if window_center is None or window_width is None:
-                # If no windowing info, use the full dynamic range
-                window_min = pixel_array.min()
-                window_max = pixel_array.max()
-            else:
-                # Handle possible multi-valued tags by taking the first value
-                if isinstance(window_center, pydicom.multival.MultiValue):
-                    window_center = window_center[0]
-                if isinstance(window_width, pydicom.multival.MultiValue):
-                    window_width = window_width[0]
-                
-                window_min = window_center - window_width / 2
-                window_max = window_center + window_width / 2
-
-            # Apply windowing
-            pixel_array = np.clip(pixel_array, window_min, window_max)
-
-            # Normalize to 0-255
-            if window_max > window_min:
-                pixel_array = ((pixel_array - window_min) / (window_max - window_min)) * 255.0
-            else: # Handle case where all pixels are the same
-                pixel_array.fill(128)
-                
-            # Handle Photometric Interpretation
-            photometric_interpretation = ds.get("PhotometricInterpretation", "MONOCHROME2")
-            if photometric_interpretation == "MONOCHROME1":
-                pixel_array = 255.0 - pixel_array
-
-            # Convert to 8-bit unsigned integer
-            image_8bit = pixel_array.astype(np.uint8)
-            
-            # Convert to RGB if needed
-            if len(image_8bit.shape) == 2:
-                image_data = np.stack([image_8bit] * 3, axis=-1)
-            else:
-                image_data = image_8bit
-                
-            self.image_data = image_data
-            height, width = image_data.shape[:2]
-            bytes_per_line = 3 * width
-            qimage = QtGui.QImage(image_data.data, width, height, bytes_per_line, QtGui.QImage.Format_RGB888)
-            pixmap = QtGui.QPixmap(qimage)
-            
-            self.image_item = QtWidgets.QGraphicsPixmapItem()
-            self.image_item.setZValue(0)
-            self.addItem(self.image_item)
-            self.mask_item = QtWidgets.QGraphicsPixmapItem()
-            self.mask_item.setZValue(1)
-            self.addItem(self.mask_item)
-
-            self.image_item.setPixmap(pixmap)
-            self.setSceneRect(self.image_item.boundingRect())
-        else:
-            image = Image.open(image_path)
-            if self.mainwindow.can_be_annotated:
-                image = image.convert('RGB')
-            self.image_data = np.array(image)
-            self.image_item = QtWidgets.QGraphicsPixmapItem()
-            self.image_item.setZValue(0)
-            self.addItem(self.image_item)
-            self.mask_item = QtWidgets.QGraphicsPixmapItem()
-            self.mask_item.setZValue(1)
-            self.addItem(self.mask_item)
-
-            self.image_item.setPixmap(QtGui.QPixmap(image_path))
-            self.setSceneRect(self.image_item.boundingRect())
+        self.image_data = self.mainwindow.current_label.get_img_data(to_rgb=True)
+
+        self.image_item = QtWidgets.QGraphicsPixmapItem()
+        self.image_item.setZValue(0)
+        self.addItem(self.image_item)
+        self.mask_item = QtWidgets.QGraphicsPixmapItem()
+        self.mask_item.setZValue(1)
+        self.addItem(self.mask_item)
+
+        height, width, channel = self.image_data.shape
+        bytes_per_line = channel * width
+        q_image = QtGui.QImage(self.image_data.data, width, height, bytes_per_line,
+                               QtGui.QImage.Format_RGB888)
+        self.image_item.setPixmap(QtGui.QPixmap.fromImage(q_image))
+        self.setSceneRect(self.image_item.boundingRect())
         self.change_mode_to_view()
 
     def unload_image(self):

ISAT/widgets/files_dock_widget.py

@@ -70,7 +70,7 @@ def dropEvent(self, event):
             suffixs = tuple(
                 ['{}'.format(fmt.data().decode('ascii').lower()) for fmt in QtGui.QImageReader.supportedImageFormats()])
             for f in os.listdir(dir):
-                if f.lower().endswith(suffixs):
+                if f.lower().endswith(suffixs) or f.lower().endswith('.dcm'):
                     # f = os.path.join(dir, f)
                     files.append(f)
             files = sorted(files)
@@ -107,7 +107,7 @@ def dropEvent(self, event):
 
             dir, file = os.path.split(path)
             files = []
-            if path.lower().endswith(suffixs):
+            if path.lower().endswith(suffixs) or path.lower().endswith('.dcm'):
                 files = [file]
 
             self.mainwindow.files_list = files

ISAT/widgets/mainwindow.py

@@ -40,52 +40,7 @@
 import requests
 import orjson
 import pydicom
-
-
-def get_windowed_image(ds):
-    # Apply Rescale Slope and Rescale Intercept if they exist
-    pixel_array = ds.pixel_array.astype(float)
-    rescale_slope = ds.get("RescaleSlope", 1)
-    rescale_intercept = ds.get("RescaleIntercept", 0)
-    if rescale_slope != 1 or rescale_intercept != 0:
-        pixel_array = pixel_array * rescale_slope + rescale_intercept
-
-    # Check for windowing information in the DICOM metadata
-    window_center = ds.get("WindowCenter", None)
-    window_width = ds.get("WindowWidth", None)
-
-    if window_center is None or window_width is None:
-        # If no windowing info, use the full dynamic range
-        window_min = pixel_array.min()
-        window_max = pixel_array.max()
-    else:
-        # Handle possible multi-valued tags by taking the first value
-        if isinstance(window_center, pydicom.multival.MultiValue):
-            window_center = window_center[0]
-        if isinstance(window_width, pydicom.multival.MultiValue):
-            window_width = window_width[0]
-        
-        window_min = window_center - window_width / 2
-        window_max = window_center + window_width / 2
-
-    # Apply windowing
-    pixel_array = np.clip(pixel_array, window_min, window_max)
-
-    # Normalize to 0-255
-    if window_max > window_min:
-        pixel_array = ((pixel_array - window_min) / (window_max - window_min)) * 255.0
-    else: # Handle case where all pixels are the same
-        pixel_array.fill(128)
-        
-    # Handle Photometric Interpretation
-    photometric_interpretation = ds.get("PhotometricInterpretation", "MONOCHROME2")
-    if photometric_interpretation == "MONOCHROME1":
-        pixel_array = 255.0 - pixel_array
-
-    # Convert to 8-bit unsigned integer
-    image_8bit = pixel_array.astype(np.uint8)
-    
-    return image_8bit
+from ISAT.annotation import get_windowed_image
 
 
 class QtBoxStyleProgressBar(QtWidgets.QProgressBar):
@@ -254,7 +209,7 @@ def run(self):
 
                     image_path = os.path.join(self.mainwindow.image_root, self.mainwindow.files_list[index])
 
-                    image_data = np.array(Image.open(image_path).convert('RGB'))
+                    image_data = self.mainwindow.current_label.get_img_data(to_rgb=True)
                     try:
                         features, original_size, input_size = self.sam_encoder(image_data)
                     except Exception as e:
@@ -1227,17 +1182,31 @@ def show_image(self, index: int, zoomfit: bool=True):
             self.plugin_manager_dialog.trigger_before_image_open(file_path)
 
             if file_path.lower().endswith('.dcm'):
-                ds = pydicom.dcmread(file_path)
-                image_data = Image.fromarray(get_windowed_image(ds))
                 self.can_be_annotated = True
             else:
-                image_data = Image.open(file_path)
-                png_palette = image_data.getpalette()
-                if png_palette is not None and file_path.endswith('.png'):
-                    self.statusbar.showMessage('This image might be a label image in VOC format.')
-                    self.can_be_annotated = False
-                else:
-                    self.can_be_annotated = True
+                try:
+                    image = Image.open(file_path)
+                    png_palette = image.getpalette()
+                    if png_palette is not None and file_path.endswith('.png'):
+                        self.statusbar.showMessage('This image might be a label image in VOC format.')
+                        self.can_be_annotated = False
+                    else:
+                        self.can_be_annotated = True
+                except Exception as e:
+                    # stausbar show error
+                    self.statusbar.showMessage(str(e), 5000)
+                    return
+
+            # load label
+            if self.can_be_annotated:
+                self.current_group = 1
+                _, name = os.path.split(file_path)
+                label_path = os.path.join(self.label_root, '.'.join(name.split('.')[:-1]) + '.json')
+                self.current_label = Annotation(file_path, label_path)
+                # 载入数据
+                self.current_label.load_annotation()
+                # get image data and info
+                self.current_label.get_img_data()
 
             if self.can_be_annotated:
                 self.actionPolygon.setEnabled(True)
@@ -1262,28 +1231,22 @@ def show_image(self, index: int, zoomfit: bool=True):
                 self.view.zoomfit()
 
             # 判断图像是否旋转
-            exif_info = image_data.getexif()
-            if exif_info and exif_info.get(274, 1) != 1:
-                warning_info = '这幅图像包含EXIF元数据，且图像的方向已被旋转.\n建议去除EXIF信息后再进行标注\n你可以使用[菜单栏]-[工具]-[处理exif标签]功能处理图像的旋转问题。'\
-                    if self.cfg['software']['language'] == 'zh' \
-                    else 'This image has EXIF metadata, and the image orientation is rotated.\nSuggest labeling after removing the EXIF metadata.\nYou can use the function of [Process EXIF tag] in [Tools] in [Menu bar] to deal with the problem of images.'
-                QtWidgets.QMessageBox.warning(self, 'Warning', warning_info, QtWidgets.QMessageBox.Ok)
+            if not file_path.lower().endswith('.dcm'):
+                image_data = Image.open(file_path)
+                exif_info = image_data.getexif()
+                if exif_info and exif_info.get(274, 1) != 1:
+                    warning_info = '这幅图像包含EXIF元数据，且图像的方向已被旋转.\n建议去除EXIF信息后再进行标注\n你可以使用[菜单栏]-[工具]-[处理exif标签]功能处理图像的旋转问题。'\
+                        if self.cfg['software']['language'] == 'zh' \
+                        else 'This image has EXIF metadata, and the image orientation is rotated.\nSuggest labeling after removing the EXIF metadata.\nYou can use the function of [Process EXIF tag] in [Tools] in [Menu bar] to deal with the problem of images.'
+                    QtWidgets.QMessageBox.warning(self, 'Warning', warning_info, QtWidgets.QMessageBox.Ok)
 
             if self.use_segment_anything and self.can_be_annotated:
                 self.segany.reset_image()
                 self.seganythread.index = index
                 self.seganythread.start()
                 self.SeganyEnabled()
 
-            # load label
             if self.can_be_annotated:
-                self.current_group = 1
-                _, name = os.path.split(file_path)
-                label_path = os.path.join(self.label_root, '.'.join(name.split('.')[:-1]) + '.json')
-                self.current_label = Annotation(file_path, label_path)
-                # 载入数据
-                self.current_label.load_annotation()
-
                 for object in self.current_label.objects:
                     try:
                         group = int(object.group)