main.tex

% Gemini theme
% See: https://rev.cs.uchicago.edu/k4rtik/gemini-uccs
% A fork of https://github.com/anishathalye/gemini

\documentclass[final]{beamer}

% ====================
% Packages
% ====================

\usepackage[size=custom,width=42,height=36,scale=0.5]{beamerposter}
\usetheme{gemini}
\usecolortheme{stanford}
\usepackage{graphicx}
\usepackage{booktabs}
\usepackage{tikz}
\usetikzlibrary{shapes, arrows, positioning}
\usepackage{pgfplots}
\pgfplotsset{compat=1.17}
\usepackage{fontspec}
\usepackage{exscale}
\usepackage{tabularx}
\usepackage{colortbl} % For \rowcolor
\usepackage{xcolor}   % For color definitions
\usepackage{tcolorbox}
\usepackage[backend=biber,style=numeric,maxnames=3,minnames=1]{biblatex}
\addbibresource{poster.bib} % Adjust the filename as needed
\usepackage{amssymb}

\setbeamercolor{headline}{bg=cardinalred,fg=white}
\setbeamercolor{footline}{bg=footergray,fg=darkcardinal}

% ====================
% Lengths
% ====================

% If you have N columns, choose \sepwidth and \colwidth such that
% (N+1)*\sepwidth + N*\colwidth = \paperwidth
\newlength{\sepwidth}
\newlength{\colwidth}
\setlength{\sepwidth}{0.025\paperwidth}
\setlength{\colwidth}{0.3\paperwidth}

\newcommand{\separatorcolumn}{\begin{column}{\sepwidth}\end{column}}


\title{Neonatal brain age models in preterm infants}


\author{\fontsize{16}{18}\selectfont Howard Chiu \inst{1} \and Adam Richie-Halford \inst{2} \and Rocio Velasco Poblaciones \inst{2} \and Melissa Scala \inst{3} \and Molly Lazarus \inst{4} \and \newline Virginia Marchman \inst{2} \and Katherine Travis \inst{4} \and Heidi Feldman \inst{2} \and Jason Yeatman \inst{1}}

\institute[shortinst]{%
  \inst{1} Graduate School of Education, Stanford University, Stanford, CA, USA \\ 
  \inst{2} Division of Developmental Behavioral Pediatrics, Stanford University School of Medicine, Stanford, CA, USA \\ 
  \inst{3} Department of Pediatrics, Division of Neonatology, Stanford University, Stanford, CA, USA \\ 
  \inst{4} Burke-Cornell Medical Research Institute, Department of Pediatrics, Weill Medical College, Cornell University, New York, NY, USA
}

% ====================
% Footer (optional)
% ====================

\footercontent{
  \href{chiuhoward.github.io}{chiuhoward.github.io} \hfill
  Thank you to all the families and healthcare professionals that made this research possible! \hfill
  \href{mailto:howardchiu@stanford.edu}{howardchiu@stanford.edu}}
% (can be left out to remove footer)

% ====================
% Logo (optional)
% ====================

% use this to include logos on the left and/or right side of the header:
% \logoright{\includegraphics[height=7cm]{logos/cs-logo-maroon.png}}
% \logoleft{\includegraphics[height=7cm]{logos/cs-logo-maroon.png}}

% ====================
% Body
% ====================

\begin{document}

% This adds the Logos on the top left and top right
\addtobeamertemplate{headline}{}
{
  \begin{tikzpicture}[remember picture,overlay]
    \node [anchor=north east, inner sep=3cm] at ([xshift=2.25cm,yshift=1.5cm]current page.north east)
    {\includegraphics[width=6.0cm]{stanford_logos/GSE-hor-logo-white.png}};
  \end{tikzpicture}

  \begin{tikzpicture}[remember picture,overlay]
    \node [anchor=north east, inner sep=3cm] at ([xshift=0.75cm,yshift=-0.5cm]current page.north east)
    {\includegraphics[width=4.25cm]{stanford_logos/SOM_DeptOfPediatrics_white_h_LG.png}};
  \end{tikzpicture}
}

\begin{frame}[t]
\begin{columns}[t]
\separatorcolumn

\begin{column}{\colwidth}
  \vspace{-15pt}
  \begin{block}{Introduction}
    Preterm birth (\textless 37 weeks of pregnancy) is an important public health issue affecting $>10\%$ of children worldwide. Among children born very preterm, \textless 32 weeks gestational age (GA), about $\frac{1}{2}$ have disrupted neurodevelopment, including abnormalities in the white matter (WM).
    
     In this study, we aim to develop a brain age model that can be used as a summary statistic of infant neurodevelopment which can guide clinical decision-making.

    \begin{columns}[t] % Ensure top alignment
      % Left column with the TikZ picture
      \begin{column}{0.45\textwidth}
        \vspace{5pt}
        \begin{tikzpicture}
          % Actual age bar (longer bar)
          \filldraw[fill=blue!30, draw=none] (0,0) rectangle (6,1);
          
          % Predicted age bar (shorter bar)
          \filldraw[fill=orange!60, draw=none] (0,0) rectangle (4,1);
          
          % Text annotations
          \node at (2,0.5) {Predicted Age};
          \node at (3,1.3) {Actual Age};
          
          % Arrow showing the brain age gap
          \draw[<->,thick] (4.2, 0.7) -- (5.8, 0.7) node[midway, below] {Gap};
          
          % Additional arrows
          \draw[<-,thick] (0.2, 1.3) -- (1.8, 1.3);
          \draw[->,thick] (4.2, 1.3) -- (5.8, 1.3);
        \end{tikzpicture}
      \end{column}
    
      % Right column with the colored text box
      \begin{column}{0.5\textwidth}
        \vspace{-20pt} % Adjust this value to vertically align
        \begin{tcolorbox}[colback=blue!10!white, colframe=blue!40!black, width=\linewidth]
          {\RaggedRight{Can we conceptualize a \\ positive brain age gap as a distal summary statistic for \textbf{poorer} infant health?}}
        \end{tcolorbox}
      \end{column}
    \end{columns}
  \end{block}

  \vspace{-5pt}

  \begin{block}{Methods}
    \begin{tikzpicture}[node distance=3cm, auto]
        % Define block styles
        \tikzstyle{block} = [rectangle, draw, fill=cardinalred, 
        text width=6em, text centered, rounded corners, minimum height=3em, font=\small, text=white]
        \tikzstyle{line} = [draw, -latex']

        % Nodes
        \node [block] (dmridata) {Raw diffusion MRI data};
        \node [block, right of=dmridata, xshift = 0.25cm] (qsiprep) {Preprocessing in QSIprep$^1$};
        \node [block, right of=qsiprep, xshift = 0.25cm] (pyAFQ) {Reconstruction \\ and \\ segmentation of whole brain tractogram in pyAFQ$^3$};
        \node [block, right of=pyAFQ, xshift = 0.25cm] (AFQInsight) {Models trained on tract profiles and target features in AFQ-Insight$^5$};

        % Lines
        \path [line] (dmridata) -- (qsiprep);
        \path [line] (qsiprep) -- (pyAFQ);
        \path [line] (pyAFQ) -- (AFQInsight);
    \end{tikzpicture}
    
    \vspace{-10pt}
    
    \begin{itemize}
        \item 3 Models
        \begin{itemize}
            \normalsize % Set the font size to normal for sub-items
            \item Principal components regression (PCR) - No regularization
            \item Lasso principal components regression (PCR-Lasso) - L1 regularization enforces sparsity
            \item Bundle-wise principal components regression with sparse group lasso penalties (PCR-SGL) - L1 and L2 regularization
        \end{itemize}
        \vspace{10pt}
        \item 3 Targets
        \begin{itemize}
            \normalsize % Set the font size to normal for sub-items
            \item GA (age since conception)
            \item Chronological age (CA; age since delivery)
            \item Post-menstrual age (PMA) at scan (sum of GA + CA)
        \end{itemize}
    \end{itemize}

    \vspace{10pt}

        \begin{tikzpicture}
        % Define colors and dimensions
        \def\barHeight{0.8}         % Height of each bar
        \def\barLengthTotal{5}       % Total length of each bar (representing 100% of data)
        \def\testBarLength{1}        % Length of test set (20%)
        \def\trainBarLength{4}       % Length of training set (80%)
        \def\hOffset{0.7}            % Horizontal displacement between splits
        \def\vOffset{0.4}            % Vertical displacement between splits
        \def\nSplits{5}              % Number of splits
        \def\offset{0.4}             % Text offset for better centering
    
        % Loop through and draw the 5 displaced bars for each split
        \foreach \i in {1,...,\nSplits} {
            % Calculate vertical and horizontal shift for each bar
            \pgfmathsetmacro{\xShift}{(\i-1)*\hOffset}
            \pgfmathsetmacro{\yShift}{(\i-1)*\vOffset}
            \pgfmathsetmacro{\testStart}{(\i-1)*\testBarLength} % Test set shift for highlighting
            
            % Total bar length
            \draw[fill=none, draw=black] 
            (\xShift, \yShift) rectangle (\xShift+\barLengthTotal, \yShift+\barHeight);
    
            % Highlight the test set section (20% of data)
            \filldraw[fill=blue!40, draw=black] 
            (\xShift+\testStart, \yShift) rectangle (\xShift+\testStart+\testBarLength, \yShift+\barHeight);
            
            % Fill the remaining part of the bar for the training set
            \filldraw[fill=orange!60, draw=black] 
            (\xShift, \yShift) rectangle (\xShift+\testStart, \yShift+\barHeight);
            \filldraw[fill=orange!60, draw=black] 
            (\xShift+\testStart+\testBarLength, \yShift) rectangle (\xShift+\barLengthTotal, \yShift+\barHeight);
        }

        % Text annotations with smaller font
        \node[font=\small] at (\xShift+\barLengthTotal, \yShift+\barHeight+1.25) {Train (80\% of data)};
        \node[font=\small] at (\xShift+\barLengthTotal+2.3, \yShift+\barHeight+1.25) {Test};
            
        % Repetition arrow
        \draw[->, thick] (\xShift+\barLengthTotal+2.5, \yShift+\barHeight) -- ++(5.5,0)
        node[midway, above] {Repeat nested cross-validation};
    
    \end{tikzpicture}


    \vspace{10pt}
    
    \begin{tikzpicture}[node distance=3cm, auto]
        % Define block styles
        \tikzstyle{block} = [rectangle, draw, fill=paloaltogreen, 
        text width=6em, text centered, rounded corners, minimum height=3em, font=\small, text=white]
        \tikzstyle{line} = [draw, -latex']

        % Nodes
        \node [block] (rawdata) {Raw data \\ available with T1, pe0 and pe1 \\ (\textit{n} = 359)};
        \node [block, right of=rawdata, xshift = 0.25cm] (qsiprep-pe1) {Completed QSIprep \\ (\textit{n} = 316)};
        \node [block, right of=qsiprep-pe1, xshift = 0.25cm] (pyAFQ-pe1) {Completed pyAFQ \\ (\textit{n} = 287)};
        \node [block, right of=pyAFQ-pe1, xshift = 0.25cm] (AFQInsight-pe1) {WM features included in brain age model after passing QC \\ (\textit{n} = 184)};

        % Lines
        \path [line] (rawdata) -- (qsiprep-pe1);
        \path [line] (qsiprep-pe1) -- (pyAFQ-pe1);
        \path [line] (pyAFQ-pe1) -- (AFQInsight-pe1);
    \end{tikzpicture}

    \vspace{-2pt}
    Their health acuity was evaluated based on a sum of binary indicators for 4 major comorbidities of prematurity (mean = 0.69, range: 0 -- 4):
    \vspace{-5pt}
    \begin{itemize}
    \item Sepsis 
    \item Necrotizing enterocolitis 
    \item Intraventricular hemorrhages (IVH) of grade 1 and above 
    \item Bronchopulmonary dysplasia of grade 2 and above 
    \end{itemize}
      
  \end{block}

\end{column}

\separatorcolumn

\begin{column}{\colwidth}
  \vspace{-15pt}
  \begin{block}{Participants (\textit{n} = 184; males = 104) were infants born at \textless 32 weeks GA. CA and GA were highly negatively correlated (\textit{r} = -0.85).}
    \vspace{-5pt}
    \begin{figure}[ht]
      \centering
      \includegraphics[clip, width=0.8\textwidth]{age_at_scan.jpg}
      \label{fig:age_at_scan}
    \end{figure}
  \end{block}
  \vspace{-20pt}
  
  \begin{block}{Right anterior thalamic radiation (ATR) microstructure contributes the most to the prediction of brain age in this sample.}
    The brain age gaps from both PCR-Lasso and SGL models were highly correlated ($\textit{r} = 0.872$). PCR-SGL produces a much sparser distribution of coefficients compared to PCR-Lasso.

    \vspace{-10pt}
    
    \begin{figure}
      \centering
      \includegraphics[clip, width=0.8\textwidth]{sgl_both.jpg}
    \end{figure}
    \vspace{-20pt}  % Adjust the value to control the space

    
    \begin{columns}[t] % Ensure top alignment
        \begin{column}{0.53\textwidth}
            \vspace{-10pt}
            \begin{figure} % sub-17055, sub-17044
            \centering  \includegraphics[clip,width=\textwidth]{brain_2.png}
            \end{figure}
      \end{column}
    
      % Right column with the colored text box
      \begin{column}{0.46\textwidth}
        \vspace{-10pt}
        \begin{tcolorbox}[colback=gray!50, colframe=cardinalred, width=\linewidth]
          {\fontsize{9.5}{10}\selectfont
          \begin{itemize}
            \item \textcolor[rgb]{1.0, 1.0, 0.4}{\rule{0.8cm}{0.3cm}} Left Arcuate (ARC)
            \item \textcolor[rgb]{0.2, 0.4, 0.6}{\rule{0.8cm}{0.3cm}} Left Anterior Thalamic Radiation (ATR) 
            \item \textcolor[rgb]{1.0, 0.4, 0.0}{\rule{0.8cm}{0.3cm}} Left Superior Longitudinal Fasciculus (SLF) 
            \item \textcolor[rgb]{0.0, 0.6, 0.2}{\rule{0.8cm}{0.3cm}} Left Uncinate (UNC)
          \end{itemize}
          Left hemisphere tracts plotted for illustrative purposes. Tract endpoints have been clipped in pyAFQ.}
        \end{tcolorbox}
      \end{column}
    \end{columns}
    
  \end{block}

\end{column}

\separatorcolumn

\begin{column}{\colwidth}
  \vspace{-15pt}
  \begin{block}{Infant WM features explain 21\% of variance in PMA at scan. Including both FA and MD increases test $R^2$.}

    In general, the infants with the highest PMA at scan had higher FA and lower MD values. 
    
    When PMA at scan is decomposed into GA at birth and CA, diffusion properties explain 15\% of variance in CA and 2\% of variance in GA, respectively (not shown in table below). 
    \begin{table}[ht]
      \centering
      \fontsize{9}{11}\selectfont
      \begin{tabularx}{\textwidth}{l c c c c c c}
        \toprule
        \textbf{Model} & \textbf{Target} & \textbf{Train $R^2$} & \textbf{Test $R^2$} & \textbf{Train MAE} &
        \textbf{Test MAE} \\
        \midrule
         PCR & PMA at scan & 0.559 & 0.198 & 8.14 & 9.84 \\
        \midrule
        \rowcolor{yellow!50} PCR-Lasso & PMA at scan & 0.491 & 0.208 & 6.99 & 7.67 \\
        PCR-Lasso (FA only) & PMA at scan & 0.428 & 0.144 & 6.75 & 7.43 \\
        PCR-Lasso (MD only) & PMA at scan & 0.180 & 0.047 & 8.30 & 8.71 \\
        \midrule
        \rowcolor{yellow!50} PCR-SGL & PMA at scan & 0.264 & 0.104 & 8.05 & 8.19 \\
        PCR-SGL (FA only) & PMA at scan & 0.220 & 0.054 & 8.14 & 8.25 \\
        PCR-SGL (MD only) & PMA at scan & 0.170 & 0.041 & 8.47 & 8.82 \\
        \bottomrule
      \end{tabularx}
    \end{table}
  \end{block}

  \vspace{-15pt}
  
  \begin{block}{Brain age gap explains additional 2\% of variance in health acuity above and beyond CA and GA. \\ Variance explained by brain age gap is greater than any WM feature alone (mean tract FA or MD).}
  
    \vspace{-25pt}

    \[    
    \textbf{Model 2:} \quad \mathrm{health\_acuity} \sim \mathrm{1} + \mathrm{CA} + \mathrm{GA} + \mathrm{brain\_age\_gap}
    \]

    \vspace{-15pt}

       \begin{table}[htbp]
        \centering
        \fontsize{12}{14}\selectfont
        \begin{tabularx}{\textwidth}{X c c c}
        \toprule
        \textbf{Variables} & \textbf{Model 1} & \textbf{Model 2} & \textbf{Model 3} \\
        \midrule
        Intercept & -0.280 & 2.93 & \checkmark \\
        CA & 0.018*** & 0.005 & \checkmark \\
        GA & 0.001 & -0.013 & \checkmark \\
        PMA at Scan Gap & - & 0.019** & - \\
        Mean Tract FA or MD & - & - & \checkmark \\
        \midrule
        \textbf{Likelihood Ratio (LR)} & - & 6.91** & \\
        \bottomrule
        \end{tabularx}
      \end{table}
      \vspace{-10pt}
     \caption{\fontsize{10}{12}\selectfont Asterisks denote level of significance: *** $\textit{p} < .001$, ** $\textit{p} < .01$, * $\textit{p} < .05$}
    \end{block}
    
    \vspace{-10pt}  % Adjust the value to control the space

   \begin{tcolorbox}[colback=blue!10!white, colframe=blue!40!black, width=\linewidth, title=\centering \textbf{Conclusion}]
    The brain age gap \textbf{shows promise} as a summary statistic for infant health, above and beyond information that can be gleaned from age and individual measures of WM microstructure.
    \end{tcolorbox}
        
    \begin{block}{References}
        \nocite{*}
        \renewcommand{\bibfont}{\fontsize{6}{7}\selectfont} % Use \footnotesize, \small, \scriptsize, or \tiny as needed
        \printbibliography
    \end{block}
  
\end{column}

\separatorcolumn
\end{columns}
\end{frame}

\end{document}