diff --git a/Cargo.toml b/Cargo.toml
index bf59417..bb87e10 100644
--- a/Cargo.toml
+++ b/Cargo.toml
@@ -12,5 +12,6 @@ repository = "https://github.com/kitsuyui/sxd_html_table"
 
 [dependencies]
 csv = "1.1.6"
+sxd-document = "0.3.2"
 sxd-xpath = "0.4.2"
 sxd_html = "0.1.1"
diff --git a/src/lib.rs b/src/lib.rs
index ea676bb..94e6a08 100644
--- a/src/lib.rs
+++ b/src/lib.rs
@@ -68,7 +68,7 @@ impl Table {
     }
 }
 
-pub fn extract_tables_from_document(html: &str) -> Result<Vec<Table>, Error> {
+pub fn extract_table_texts_from_document(html: &str) -> Result<Vec<Table>, Error> {
     let package = sxd_html::parse_html(html);
     let document = package.as_document();
     #[allow(clippy::expect_used)]
@@ -79,7 +79,26 @@ pub fn extract_tables_from_document(html: &str) -> Result<Vec<Table>, Error> {
     };
     let mut tables = vec![];
     for node in table_nodes.document_order() {
-        match extract_table(&node) {
+        match extract_table_texts(&node) {
+            Ok(table) => tables.push(table),
+            Err(e) => return Err(e),
+        }
+    }
+    Ok(tables)
+}
+
+pub fn extract_table_elements_from_document(html: &str) -> Result<Vec<Table>, Error> {
+    let package = sxd_html::parse_html(html);
+    let document = package.as_document();
+    #[allow(clippy::expect_used)]
+    let val = evaluate_xpath_node(document.root(), "//table").expect("XPath evaluation failed");
+
+    let Value::Nodeset(table_nodes) = val else {
+        panic!("Expected node set");
+    };
+    let mut tables = vec![];
+    for node in table_nodes.document_order() {
+        match extract_table_elements(&node) {
             Ok(table) => tables.push(table),
             Err(e) => return Err(e),
         }
@@ -90,26 +109,25 @@ pub fn extract_tables_from_document(html: &str) -> Result<Vec<Table>, Error> {
 fn extract_rowspan_and_colspan(node: &Node) -> (usize, usize) {
     #[allow(clippy::expect_used)]
     let element = node.element().expect("Expected element");
-    let rowspan = element
-        .attribute_value("rowspan")
-        .unwrap_or("1")
-        .parse::<usize>()
-        .unwrap_or(1);
-    let colspan = element
-        .attribute_value("colspan")
+    let rowspan = extract_span(element, "rowspan");
+    let colspan = extract_span(element, "colspan");
+    (rowspan, colspan)
+}
+
+fn extract_span(element: sxd_document::dom::Element, name: &str) -> usize {
+    element
+        .attribute_value(name)
         .unwrap_or("1")
         .parse::<usize>()
-        .unwrap_or(1);
-    (rowspan, colspan)
+        .unwrap_or(1)
 }
 
-fn extract_table(node: &Node) -> Result<Table, Error> {
+pub fn map_table_cell(node: &Node, f: fn(&Node) -> String) -> Result<Table, Error> {
     let tr_nodes = match evaluate_xpath_node(*node, "./tbody/tr") {
         Ok(Value::Nodeset(tr_nodes)) => tr_nodes,
         _ => return Err(Error::InvalidDocument),
     };
     let tr_nodes = tr_nodes.document_order();
-
     let mut map: HashMap<(usize, usize), String> = HashMap::new();
     let mut header_map: HashMap<(usize, usize), bool> = HashMap::new();
     for (row_index, tr) in tr_nodes.iter().enumerate() {
@@ -121,7 +139,7 @@ fn extract_table(node: &Node) -> Result<Table, Error> {
         let mut col_index = 0;
         for (_, cell_node) in cell_nodes.iter().enumerate() {
             let (row_size, col_size) = extract_rowspan_and_colspan(cell_node);
-            let text = &cell_node.string_value();
+            let text = f(cell_node);
             #[allow(clippy::expect_used)]
             let is_header = cell_node.element().expect("Expected element").name() == "th".into();
             while map.contains_key(&(row_index, col_index)) {
@@ -135,16 +153,42 @@ fn extract_table(node: &Node) -> Result<Table, Error> {
             }
         }
     }
+    Ok(map_to_table(&map))
+}
+
+fn map_to_table(map: &HashMap<(usize, usize), String>) -> Table {
     let rows = map.keys().map(|(i, _)| i).max().unwrap_or(&0) + 1;
     let cols = map.keys().map(|(_, j)| j).max().unwrap_or(&0) + 1;
     let mut table = Table::new((rows, cols));
     for ((i, j), text) in map {
-        table.cells[i * table.size.1 + j] = Some(text);
+        table.cells[i * table.size.1 + j] = Some(text.to_string());
     }
-    for ((i, j), is_header) in header_map {
-        table.headers[i * table.size.1 + j] = is_header;
+    table
+}
+
+fn extract_table_texts(node: &Node) -> Result<Table, Error> {
+    map_table_cell(node, |node| node.string_value())
+}
+
+fn extract_table_elements(node: &Node) -> Result<Table, Error> {
+    map_table_cell(node, |node| element_to_html(node))
+}
+
+fn element_to_html(node: &Node) -> String {
+    let mut buf = Vec::new();
+    let package = sxd_document::Package::new();
+    let doc = package.as_document();
+    let root = doc.root();
+    match node.element() {
+        Some(element) => {
+            root.append_child(element.clone());
+        }
+        None => (),
     }
-    Ok(table)
+    #[allow(clippy::expect_used)]
+    sxd_document::writer::format_document(&doc, &mut buf).expect("Failed to format document");
+    #[allow(clippy::expect_used)]
+    String::from_utf8(buf).expect("Failed to convert to UTF-8")
 }
 
 fn evaluate_xpath_node<'d>(
@@ -179,7 +223,7 @@ mod tests {
             </body>
         </html>
         "#;
-        let result = extract_tables_from_document(html).unwrap();
+        let result = extract_table_texts_from_document(html).unwrap();
         assert_eq!(result.len(), 1);
         assert_eq!(result[0].to_csv().unwrap(), "1,2\n");
 
@@ -202,7 +246,7 @@ mod tests {
             </body>
         </html>
         "#;
-        let result = extract_tables_from_document(html).unwrap();
+        let result = extract_table_texts_from_document(html).unwrap();
         assert_eq!(result.len(), 2);
         assert_eq!(result[0].to_csv().unwrap(), "1,2\n",);
         assert_eq!(result[1].to_csv().unwrap(), "3,4\n",);
@@ -218,12 +262,12 @@ mod tests {
             </body>
         </html>
         "#;
-        let result = extract_tables_from_document(html).unwrap();
+        let result = extract_table_texts_from_document(html).unwrap();
         assert_eq!(result.len(), 0);
 
         // empty html
         let html = r#""#;
-        let result = extract_tables_from_document(html).unwrap();
+        let result = extract_table_texts_from_document(html).unwrap();
         assert_eq!(result.len(), 0);
     }
 
@@ -245,7 +289,7 @@ mod tests {
             </body>
         </html>
         "#;
-        let result = extract_tables_from_document(html).unwrap();
+        let result = extract_table_texts_from_document(html).unwrap();
         assert_eq!(result.len(), 1);
         assert_eq!(result[0].to_csv().unwrap(), "1,2\n3,4\n");
     }
@@ -267,7 +311,7 @@ mod tests {
             </body>
         </html>
         "#;
-        let result = extract_tables_from_document(html).unwrap();
+        let result = extract_table_texts_from_document(html).unwrap();
         assert_eq!(result.len(), 1);
         assert_eq!(result[0].to_csv().unwrap(), "A,B\nA,C\n");
 
@@ -288,7 +332,7 @@ mod tests {
             </body>
         </html>
         "#;
-        let result = extract_tables_from_document(html).unwrap();
+        let result = extract_table_texts_from_document(html).unwrap();
         assert_eq!(result.len(), 1);
         assert_eq!(result[0].to_csv().unwrap(), "A,A,B\nC,D,E\n");
 
@@ -333,7 +377,7 @@ mod tests {
         </html>
         "#;
 
-        let result = extract_tables_from_document(html).unwrap();
+        let result = extract_table_texts_from_document(html).unwrap();
         assert_eq!(result.len(), 6);
         assert_eq!(result[0].to_csv().unwrap(), "A,A,B\nA,A,C\n");
         assert_eq!(result[1].to_csv().unwrap(), "a,b,c\nd,e,f\n");