/** * REST client for the local FastAPI inference engine (/predict, /health). * When the engine is unreachable, the Detector Studio falls back to a * clearly-labeled simulated demo mode so the UI remains fully explorable. */ export const API_BASE: string = (import.meta.env["VITE_API_BASE"] as string | undefined) ?? "http://localhost:8000"; export type Health = { status: string; model_loaded: boolean; device: string; model_name: string; }; export type Prediction = { class_name: string; confidence: number; probabilities: Record; inference_time_ms: number; }; /** * Error thrown when the backend received the request but rejected it * (e.g. color image on a grayscale-only pipeline, corrupt file). * Distinct from a network failure so the UI can surface it instead of * silently falling back to simulations. */ export class ApiError extends Error { status: number; constructor(message: string, status: number) { super(message); this.name = "ApiError"; this.status = status; } } export async function checkHealth(timeoutMs = 3000): Promise { const controller = new AbortController(); const timer = setTimeout(() => controller.abort(), timeoutMs); try { const res = await fetch(`${API_BASE}/health`, { signal: controller.signal }); if (!res.ok) throw new Error(`Health check failed: ${res.status}`); return (await res.json()) as Health; } finally { clearTimeout(timer); } } export async function runTumorInference(imageFile: File): Promise { const formData = new FormData(); formData.append("file", imageFile); const res = await fetch(`${API_BASE}/predict`, { method: "POST", body: formData }); if (!res.ok) { let detail = "Diagnostic scan failed"; try { const err = await res.json(); // The API's exception handler returns the message in `error`, // with `detail` as an empty companion field. detail = err.error ?? err.detail ?? detail; } catch { /* non-JSON error body */ } throw new ApiError(detail, res.status); } return (await res.json()) as Prediction; } /* ---------- Simulated demo mode (engine offline) ---------- */ function hashString(input: string): number { let h = 2166136261; for (let i = 0; i < input.length; i++) { h ^= input.charCodeAt(i); h = Math.imul(h, 16777619); } return h >>> 0; } /** * Deterministic, plausible-looking result derived from the file name so the * demo is reproducible while still showing all four classes across uploads. */ export function simulatePrediction(file: File): Prediction { const h = hashString(file.name + file.size); const pick = h % 100; const className = pick < 40 ? "Glioma" : pick < 63 ? "Meningioma" : pick < 82 ? "Pituitary" : "No Tumor"; const confidence = 0.92 + ((h >>> 8) % 700) / 10000; // 0.92 - 0.9899 const remaining = 1 - confidence; const others = ["Glioma", "Meningioma", "No Tumor", "Pituitary"].filter((c) => c !== className); const weights = [((h >>> 12) % 45) + 30, ((h >>> 16) % 35) + 15, ((h >>> 20) % 25) + 5]; const wSum = weights.reduce((a, b) => a + b, 0); const probabilities: Record = { [className]: confidence }; others.forEach((c, i) => { const w = weights[i] ?? 10; probabilities[c] = Number(((remaining * w) / wSum).toFixed(4)); }); return { class_name: className, confidence: Number(confidence.toFixed(4)), probabilities, inference_time_ms: Number((10.2 + ((h >>> 4) % 40) / 10).toFixed(1)), // 10.2 - 14.1 ms }; } /** Seeded Grad-CAM focus point (percentages) for the demo heatmap overlay. */ export function gradcamFocus(file: File): { x: number; y: number } { const h = hashString(file.name); return { x: 28 + (h % 44), y: 26 + ((h >>> 9) % 40) }; }