info.json

{
  "system_prompt": "You are MedGemma, an expert medical AI assistant specialized in analyzing medical images and generating accurate, insightful radiological reports. You provide detailed, professional descriptions of findings, always basing your analysis on visible evidence in the image. Be concise yet comprehensive, use medical terminology appropriately, and structure your response as a clear radiological report. If asked about diagnoses, emphasize that you are not a substitute for professional medical advice.",
  "classification_brain_tumor_message": "Analyze the provided brain MRI image, which has been classified using Grad-CAM visualization. On the left is the original image; on the right is the Grad-CAM heat map highlighting relevant features. In the legend, the class probabilities for glioma, meningioma, no_tumor, and pituitary are shown. Provide a detailed radiological report on the tumor characteristics, supporting features for the predicted class, and clinical implications based on the highest probability class.",
  "classification_chest_xray_message": "Analyze the provided chest X-ray image, which has been classified using Grad-CAM visualization. On the left is the original image; on the right is the Grad-CAM heat map highlighting relevant features. In the legend, the class probabilities for COVID19, NORMAL, PNEUMONIA, and TUBERCULOSIS are shown. Provide a detailed radiological report on lung findings, opacities, consolidations, or other abnormalities, and discuss implications based on the predicted class with highest probability.",
  "classification_lung_histopathology_message": "Analyze the provided lung tissue slide image, which has been classified using Grad-CAM visualization. On the left is the original image; on the right is the Grad-CAM heat map highlighting relevant cellular features. In the legend, the class probabilities for adenocarcinoma, benign, and squamous_cell_carcinoma are shown. Provide a detailed histopathological report on cellular morphology, glandular structures, or keratinization, and interpret findings based on the predicted class with highest probability.",
  "detection_blood_message": "Analyze the provided blood smear image, which has been processed for object detection. Bounding boxes are drawn around detected cells with labels (Basophil, Eosinophil, Lymphocyte, Monocyte, Neutrophil, Platelets, RBC) and confidence probabilities. Provide a detailed hematological report on cell morphology, distribution, counts, and any abnormalities observed, summarizing potential insights into blood composition.",
  "detection_breast_cancer_message": "Analyze the provided mammogram or ultrasound image, which has been processed for object detection. A bounding box is drawn around the detected lesion with the label 'breast_cancer' and confidence probability. Provide a detailed radiological report on the lesion's shape, margins, density, surrounding tissue changes, and features suggestive of malignancy, including recommendations for further evaluation.",
  "detection_fracture_message": "Analyze the provided X-ray image, which has been processed for object detection. A bounding box is drawn around the detected fracture site with the label 'fracture' and confidence probability. Provide a detailed radiological report on the fracture line, displacement, alignment, associated soft tissue changes, fracture type classification, and clinical implications.",
  "segmentation_brats_single_message": "Analyze this single brain MRI slice with overlaid segmentation. Segmentation colors indicate: - RED: Non-Enhancing Tumor Core (NCR/NET) - necrotic tumor tissue - GREEN: Peritumoral Edema (ED) - swelling around the tumor - BLUE: Enhancing Tumor (ET) - actively growing tumor tissue. Provide a detailed radiological report describing tumor regions, their characteristics, necrosis, edema, enhancement, and overall lesion assessment.",
  "segmentation_brats_three_message": "Analyze these 3 consecutive brain MRI slices (10-slice intervals) with overlaid segmentation. Segmentation colors indicate: - RED: Non-Enhancing Tumor Core (NCR/NET) - necrotic tumor tissue - GREEN: Peritumoral Edema (ED) - swelling around the tumor - BLUE: Enhancing Tumor (ET) - actively growing tumor tissue. Provide a detailed radiological report on tumor progression, spatial distribution of components across slices, and evolution of the lesion.",
  "segmentation_brats_five_message": "Analyze these 5 consecutive brain MRI slices (5-slice intervals) with overlaid segmentation. Segmentation colors indicate: - RED: Non-Enhancing Tumor Core (NCR/NET) - necrotic tumor tissue - GREEN: Peritumoral Edema (ED) - swelling around the tumor - BLUE: Enhancing Tumor (ET) - actively growing tumor tissue. Provide a comprehensive radiological report on the complete tumor profile, extent, necrosis, edema, enhancing regions, and volumetric changes."
}