example.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# 🧠 NOVA Benchmark: Extreme Stress-Test for Out-of-Distribution Detection in Brain MRI\n",
    "\n",
    "Welcome to the NOVA dataset — a carefully curated, evaluation-only benchmark designed to push the limits of machine learning models in real-world clinical scenarios. With over **900 brain MRI scans**, **281 rare pathologies**, and **rich clinical metadata**, NOVA goes beyond traditional anomaly detection.\n",
    "\n",
    "This notebook walks you through how to:\n",
    "\n",
    "- Load the NOVA dataset directly from Hugging Face 🤗\n",
    "- Access images, captions, and diagnostic metadata\n",
    "- Visualize expert-annotated bounding boxes (gold standard and raters)\n",
    "- Explore one of the most challenging testbeds for generalization and reasoning under uncertainty\n",
    "\n",
    "> ⚠️ This benchmark is intended **only for evaluation**. No training should be performed on NOVA.\n",
    "\n",
    "📘 For more details, visit the [dataset page on Hugging Face](https://huggingface.co/datasets/Ano-2090/Nova).\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "import matplotlib\n",
    "matplotlib.use('agg')\n",
    "import matplotlib.pyplot as plt\n",
    "import matplotlib.patches as patches\n",
    "from datasets import load_dataset\n",
    "import random"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Load dataset\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "ds = load_dataset(\"Ano-2090/Nova\", trust_remote_code=True)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Select a random example\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "example = random.choice(ds[\"test\"])\n",
    "image = example[\"image\"]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Create figure and display image\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "fig, ax = plt.subplots(1, figsize=(8, 8))\n",
    "ax.imshow(image)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Plot gold standard bounding boxes (gold)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "bbox = example[\"bbox_gold\"]\n",
    "for x, y, w, h in zip(bbox[\"x\"], bbox[\"y\"], bbox[\"width\"], bbox[\"height\"]):\n",
    "    rect = patches.Rectangle((x, y), w, h, linewidth=2, edgecolor=\"gold\", facecolor=\"none\")\n",
    "    ax.add_patch(rect)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Plot rater bounding boxes (turquoise, salmon) with labels\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "colors = ['#40E0D0', '#FA8072']\n",
    "raters = example[\"bbox_raters\"]\n",
    "if raters:\n",
    "    for i in range(len(raters[\"x\"])):\n",
    "        rater = raters[\"rater\"][i]\n",
    "        x = raters[\"x\"][i]\n",
    "        y = raters[\"y\"][i]\n",
    "        w = raters[\"width\"][i]\n",
    "        h = raters[\"height\"][i]\n",
    "        rect = patches.Rectangle((x, y), w, h, linewidth=1.5, edgecolor=colors[i], facecolor=\"none\", linestyle=\"--\")\n",
    "        ax.add_patch(rect)\n",
    "        if i == 0:\n",
    "            ax.text(x, y - 5, rater, color=colors[i], fontsize=8, weight=\"bold\", va=\"bottom\")\n",
    "        else: \n",
    "            ax.text(x + w/2, y + h + 15, rater, color=colors[i], fontsize=8, weight=\"bold\", va=\"bottom\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Visualize example\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "plt.title(f'{example[\"filename\"]} — {example[\"final_diagnosis\"]}', fontsize=12)\n",
    "plt.axis(\"off\")\n",
    "plt.tight_layout()\n",
    "display(fig)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Print other metadata \n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "print('*-------------------------------------------------------*')\n",
    "print('*-------------------------------------------------------*')\n",
    "print('caption:', example[\"caption\"])\n",
    "print('*-------------------------------------------------------*')\n",
    "print('clinical history:', example[\"clinical_history\"])\n",
    "print('*-------------------------------------------------------*')\n",
    "print('differential diagnosis:', example[\"differential_diagnosis\"])\n",
    "print('*-------------------------------------------------------*')\n",
    "print('final diagnosis:', example[\"final_diagnosis\"])\n",
    "print('*-------------------------------------------------------*')\n",
    "print('*-------------------------------------------------------*')"
   ]
  }
 ],
 "metadata": {
  "language_info": {
   "name": "python"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}