"""EyeQC clinical theme + custom CSS.""" import gradio as gr THEME = gr.themes.Soft( primary_hue=gr.themes.colors.teal, secondary_hue=gr.themes.colors.cyan, neutral_hue=gr.themes.colors.slate, font=[gr.themes.GoogleFont("Inter"), "system-ui", "sans-serif"], font_mono=[gr.themes.GoogleFont("IBM Plex Mono"), "monospace"], ).set( body_background_fill="#070d11", block_background_fill="#0e1a20", block_border_width="1px", block_border_color="#1b2c34", block_radius="16px", block_label_text_color="#7fb8c4", block_title_text_color="#d7e2e6", button_primary_background_fill="linear-gradient(92deg,#0b6b7d,#22a5b8)", button_primary_background_fill_hover="linear-gradient(92deg,#0b7c90,#2bbad0)", button_primary_text_color="#ffffff", body_text_color="#cdd9dd", ) CSS = """ :root { --ink:#eaf2f4; --teal:#2aa5b8; --deep:#0b3c49; --muted:#7f97a0; --line:#1b2c34; --panel:#0e1a20; } .gradio-container { max-width: 1440px !important; margin: 0 auto !important; } #hero { background: radial-gradient(900px 240px at 88% -30%, rgba(42,165,184,.22), transparent 60%), linear-gradient(120deg, #0f4a58 0%, #0a3a45 40%, #082730 75%, #061a20 100%); border:1px solid #14343d; border-radius:22px; padding:28px 32px; margin-bottom:14px; box-shadow:0 14px 50px rgba(0,0,0,.45); } #hero .brand { display:flex; align-items:center; gap:14px; } #hero .logo { width:44px; height:44px; border-radius:12px; flex:none; background:linear-gradient(135deg,#22a5b8,#0b6b7d); box-shadow:0 6px 18px rgba(34,165,184,.4); display:flex; align-items:center; justify-content:center; font-size:24px; } #hero h1 { font-size:32px; font-weight:800; letter-spacing:-.6px; margin:0; color:#fff; } #hero .tag { font-size:12px; color:#8fd0da; letter-spacing:2px; text-transform:uppercase; margin:3px 0 0 0; font-weight:600; } #hero .sub { color:#9fc2c8; font-size:14px; margin:12px 0 0 0; max-width:980px; line-height:1.55; } #hero .pill { display:inline-block; background:rgba(42,165,184,.10); border:1px solid rgba(42,165,184,.28); color:#bfe4ea; font-size:11.5px; padding:5px 12px; border-radius:999px; margin:12px 8px 0 0; } .section-title { font-size:12px; font-weight:700; text-transform:uppercase; letter-spacing:1.4px; color:var(--teal); margin:8px 0 4px 2px; } .verdict-card { border-radius:16px; padding:18px 20px; color:#fff; font-weight:700; text-align:center; box-shadow:0 8px 24px rgba(0,0,0,.3); } .verdict-PASS { background:linear-gradient(135deg,#1f9d61,#0f6b43); } .verdict-ACCEPTABLE { background:linear-gradient(135deg,#d69e2e,#9c7012); } .verdict-FAIL { background:linear-gradient(135deg,#e05252,#a82f2f); } .verdict-card .big { font-size:36px; line-height:1; } .verdict-card .sm { font-size:12px; font-weight:500; opacity:.92; } .chip { display:inline-block; padding:6px 12px; border-radius:999px; font-size:12px; font-weight:600; margin:6px 6px 0 0; border:1px solid var(--line); } .chip-g { background:rgba(40,192,127,.14); color:#7fe3b3; border-color:rgba(40,192,127,.4);} .chip-u { background:rgba(224,82,82,.14); color:#f2a3a3; border-color:rgba(224,82,82,.4);} .chip-q { background:rgba(224,178,60,.14); color:#f0d38a; border-color:rgba(224,178,60,.4);} .reason { background:#0c161b; border:1px solid var(--line); border-left:4px solid var(--teal); border-radius:12px; padding:14px 16px; color:var(--ink); font-size:13.5px; line-height:1.55; } .reason table { width:100%; border-collapse:collapse; margin-top:6px; } .reason th { color:var(--muted); text-align:left; font-weight:600; padding:2px 6px; } .reason td { padding:2px 6px; } .footer-note { color:var(--muted); font-size:12px; text-align:center; margin-top:16px; line-height:1.6; } .md-note { color:var(--muted); font-size:12.5px; } table td, table th { font-size:12.5px !important; } """ def hero_html(): return """

EyeQC

Retinal image quality & foundation-model bench

EyeQC helps you trust what you're looking at. It checks whether a retinal photo is clear enough to read, shows you exactly where and why an image falls short, and evens out the differences between cameras and sites so a cohort is comparable. It also asks a harder question of AI models: is a diagnosis picking up real disease, or just reacting to a blurry, poorly-lit image?

Vessel-aware gradability Conformal verdicts Failure localisation ComBat harmonisation Degradation-sensitivity probe Spatial disentanglement FLAIR VQA
""" def verdict_card_html(summary): v = summary["verdict"] return f"""
{summary['composite']:.0f}/100
{v}
clinical band: {summary['band']}
""" def conformal_chip_html(pred): lab = pred["label"] cls = {"gradable": "chip-g", "ungradable": "chip-u"}.get(lab, "chip-q") cov = pred.get("coverage") cov_txt = f" · {int(cov*100)}% coverage" if cov else "" src = pred.get("source", "") return (f'conformal: {lab}{cov_txt}' f'  ({src})')