背景：

上一篇博文DICOM：DICOM万能编辑工具之Sante DICOM Editor介绍了DICOM万能编辑工具，在日常使用过程中发现，“只要Sante DICOM Editor打不开的数据，基本可以判定此DICOM文件格式错误（准确率达99.9999%^_^）”。在感叹Sante DICOM Editor神器牛掰的同时，想了解一下其底层是如何实现的。通过日常使用以及阅读软件帮助手册推断其底层依赖库很可能是dcmtk，就如同本人使用dcmtk、fo-dicom、dcm4che3等诸多DICOM开源库遇到的兼容性问题类似，——dcmtk兼容性最强，fo-dicom次之，dcm4che3最差。

问题：

本篇通过对比dcmtk3.6与dcm4che3.x解析同一特殊dicom文件（包含非标准VR的元素）分析dcmtk、dcm4che以及fo-dicom数据加载的兼容性问题。
特殊的dicom文件内容如下：
28 00 20 01 20 20 02 00 30 F8，具体描述如下：

使用dcmtk与fo-dicom加载数据时都未出现错误，例如dcmtk加载数据时的提示如下：

由此可以看出dcmtk已经顺利识别出了非标准VR的元素(0028,0120)，并成功加载。
虽然使用fo-dicom加载数据没有出现错误，但是对于上述非标准VR的元素(0028,0120)后的元素未顺利加载，如下图所示：

而dcm4che3加载过程中直接弹出了错误，如下所示：

问题分析：

出现该问题的原因是dcm4che3和fo-dicom在解析0028,0120元素时，对于20 20的非标准VR无法识别。下文中将通过分析dcm4che3与dcmtk的源码来定位问题的具体位置并给出解决方案（此处暂时只对比分析了dcm4che3.3.8最新版与dmctk3.6的源码，对于fo-dicom的源码分析待后续整理完成后再补充）。

1. dcmtk3.6源码：

使用dcmtk编写本次数据加载测试工程，简单的示例代码如下：

<span style="color:#000000"><code class="language-C"><span style="color:#abb2bf">int</span> main()
{OFLog::configure(OFLogger::TRACE_LOG_LEVEL);<span style="color:#abb2bf">char</span>* ifname = <span style="color:#abb2bf">"c:\\1.dcm"</span>;E_FileReadMode readMode = <span style="color:#abb2bf"><em>/*ERM_fileOnly*/</em></span>ERM_autoDetect;E_TransferSyntax xfer =  EXS_Unknown;Uint32 maxReadLength = DCM_MaxReadLength;<span style="color:#abb2bf">bool</span> loadIntoMemory = <span style="color:#abb2bf">true</span>;DcmFileFormat dfile;DcmObject *dset = &dfile;<span style="color:#abb2bf">if</span> (readMode == ERM_dataset) dset = dfile.getDataset();OFCondition cond = dfile.loadFile(ifname, xfer, EGL_noChange, maxReadLength, readMode);<span style="color:#abb2bf">if</span> (cond.bad()){<span style="color:#abb2bf">return</span> <span style="color:#abb2bf">1</span>;}<span style="color:#abb2bf">return</span> <span style="color:#abb2bf">0</span>;
}</code></span>

单步调试，可以知道dcmtk加载dicom文件的流程如下：

创建DcmMetaInfo、DcmDataset元素
分别加载DcmMetaInfo、DcmDataset元素
使用DcmItem中的readGroupLength、readTagAndLength、readSubElement逐步加载DcmMetaInfo、DcmDataset的各个子元素。

在DcmItem类中对于非标准VR元素有相应的警告提示信息，

<span style="color:#000000"><code class="language-C">/* <span style="color:#abb2bf">if</span> the VR which was read is not a standard VR, print a <span style="color:#abb2bf">warning</span> */<span style="color:#abb2bf">if</span> (!vr.isStandard()){OFOStringStream oss;oss << <span style="color:#abb2bf">"DcmItem: Non-standard VR '"</span><< ((OFstatic_cast(unsigned char, vrstr[<span style="color:#abb2bf">0</span>]) < <span style="color:#abb2bf">32</span>) ? <span style="color:#abb2bf">' '</span> : vrstr[<span style="color:#abb2bf">0</span>])<< ((OFstatic_cast(unsigned char, vrstr[<span style="color:#abb2bf">1</span>]) < <span style="color:#abb2bf">32</span>) ? <span style="color:#abb2bf">' '</span> : vrstr[<span style="color:#abb2bf">1</span>]) << <span style="color:#abb2bf">"' ("</span><< STD_NAMESPACE hex << STD_NAMESPACE setfill(<span style="color:#abb2bf">'0'</span>)<< STD_NAMESPACE setw(<span style="color:#abb2bf">2</span>) << OFstatic_cast(unsigned int, vrstr[<span style="color:#abb2bf">0</span>] & <span style="color:#abb2bf">0xff</span>) << <span style="color:#abb2bf">"\\"</span><< STD_NAMESPACE setw(<span style="color:#abb2bf">2</span>) << OFstatic_cast(unsigned int, vrstr[<span style="color:#abb2bf">1</span>] & <span style="color:#abb2bf">0xff</span>)<< <span style="color:#abb2bf">") encountered while parsing element "</span> << newTag << OFStringStream_ends;OFSTRINGSTREAM_GETSTR(oss, tmpString)/* encoding of this data element might be wrong, <span style="color:#abb2bf">try</span> to correct it */<span style="color:#abb2bf">if</span> (dcmAcceptUnexpectedImplicitEncoding.get()){DCMDATA_WARN(tmpString << <span style="color:#abb2bf">", trying again with Implicit VR Little Endian"</span>);/* put back read bytes to input stream <span style="color:#abb2bf">...</span> */inStream.putback();bytesRead = <span style="color:#abb2bf">0</span>;/* <span style="color:#abb2bf">...</span> and retry with Implicit VR Little Endian transfer syntax */<span style="color:#abb2bf">return</span> readTagAndLength(inStream, EXS_LittleEndianImplicit, tag, length, bytesRead);} <span style="color:#abb2bf">else</span> {DCMDATA_WARN(tmpString << <span style="color:#abb2bf">", assuming "</span> << (vr.usesExtendedLengthEncoding() ? <span style="color:#abb2bf">"4"</span> : <span style="color:#abb2bf">"2"</span>)<< <span style="color:#abb2bf">" byte length field"</span>);}OFSTRINGSTREAM_FREESTR(tmpString)}/* set the VR which was read <span style="color:#abb2bf">in</span> the above created tag object. */newTag.setVR(vr);/* increase counter by <span style="color:#abb2bf">2</span> */bytesRead += <span style="color:#abb2bf">2</span>;</code></span>

在警告后，对于非标准VR元素的处理过程如下：

<span style="color:#000000"><code class="language-C"> <span style="color:#abb2bf"><em>/* read the value in the length field. In some cases, it is 4 bytes wide, in other */</em></span><span style="color:#abb2bf"><em>/* cases only 2 bytes (see DICOM standard part 5, section 7.1.1) */</em></span><span style="color:#abb2bf">if</span> (xferSyn.isImplicitVR() || nxtobj == EVR_na)   <span style="color:#abb2bf"><em>//note that delimitation items don't have a VR</em></span>{inStream.read(&valueLength, <span style="color:#abb2bf">4</span>);            <span style="color:#abb2bf"><em>//length field is 4 bytes wide</em></span>swapIfNecessary(gLocalByteOrder, byteOrder, &valueLength, <span style="color:#abb2bf">4</span>, <span style="color:#abb2bf">4</span>);bytesRead += <span style="color:#abb2bf">4</span>;} <span style="color:#abb2bf">else</span> {                                       <span style="color:#abb2bf"><em>//the transfer syntax is explicit VR</em></span>DcmVR vr(newTag.getEVR());<span style="color:#abb2bf">if</span> (vr.usesExtendedLengthEncoding()){Uint16 reserved;inStream.read(&reserved, <span style="color:#abb2bf">2</span>);           <span style="color:#abb2bf"><em>// 2 reserved bytes</em></span>inStream.read(&valueLength, <span style="color:#abb2bf">4</span>);        <span style="color:#abb2bf"><em>// length field is 4 bytes wide</em></span>swapIfNecessary(gLocalByteOrder, byteOrder, &valueLength, <span style="color:#abb2bf">4</span>, <span style="color:#abb2bf">4</span>);bytesRead += <span style="color:#abb2bf">6</span>;} <span style="color:#abb2bf">else</span> {Uint16 tmpValueLength;inStream.read(&tmpValueLength, <span style="color:#abb2bf">2</span>);     <span style="color:#abb2bf"><em>// length field is 2 bytes wide</em></span>swapIfNecessary(gLocalByteOrder, byteOrder, &tmpValueLength, <span style="color:#abb2bf">2</span>, <span style="color:#abb2bf">2</span>);bytesRead += <span style="color:#abb2bf">2</span>;valueLength = tmpValueLength;}}</code></span>

由上述代码可知，0028,0120的VR=20,20，被dcmtk解析为 EVR_UNKNOWN2B类型，如同代码注释中所描述：

/// used internally for elements with unknown VR with 2-byte length field in explicit VR
EVR_UNKNOWN2B

DICOM标准PS5的7.1.2有对于非标准VR的相关描述，如下：

2. dcm4che3.3.8源码：

再对比dcm4che3.3.8的源码，单步调试发现，对于0028,0120的VR=20,20，被dcmtk直接标记为UN类型，

<span style="color:#000000"><code class="language-Java"><span style="color:#abb2bf">public</span> <span style="color:#abb2bf">static</span> VR <span style="color:#abb2bf">valueOf</span>(<span style="color:#abb2bf">int</span> code) {<span style="color:#abb2bf">try</span> {VR vr = VALUE_OF[indexOf(code)];<span style="color:#abb2bf">if</span> (vr != <span style="color:#abb2bf">null</span>)<span style="color:#abb2bf">return</span> vr;} <span style="color:#abb2bf">catch</span> (IndexOutOfBoundsException e) {}LOG.warn(<span style="color:#abb2bf">"Unrecogniced VR code: {0}H - treat as UN"</span>,Integer.toHexString(code));<span style="color:#abb2bf">return</span> UN;}</code></span>

并且在dcm4che3中对于UN类型定义为

此处UN类型是参照上述截图中DICOM3.0标准对于VR=UN（unknown）类型的标签约束来定义的，即，其VR字段应该是四个字节。然而此处0028,0120的VR=20,20后的Value Length只有两个字节02 00。因此导致dcm4che3在加载0028,0120元素时，将其长度错误地解析为4163895298，即十六进制的F8 30 00 02，如下图所示：

解决方案：

至此我们找到了dcm4che3错误解析0028,0120的VR=20,20非标准VR元素的原因。对于这种非标准VR不能统一当做VR.UN类型进行处理，而应该根据其后续的Value Length的具体长度为2或者4来进行分类处理（关于该问题后续博文会继续深入剖析，请注意），需要修改的地方有两处：

1. 正确解析Non-standard VR：

<span style="color:#000000"><code class="language-Java"><span style="color:#abb2bf"><em>//VR.java,Line 110</em></span>
<span style="color:#abb2bf">public</span> <span style="color:#abb2bf">static</span> VR <span style="color:#abb2bf">valueOf</span>(<span style="color:#abb2bf">int</span> code) {<span style="color:#abb2bf">try</span> {VR vr = VALUE_OF[indexOf(code)];<span style="color:#abb2bf">if</span> (vr != <span style="color:#abb2bf">null</span>)<span style="color:#abb2bf">return</span> vr;} <span style="color:#abb2bf">catch</span> (IndexOutOfBoundsException e) {}LOG.warn(<span style="color:#abb2bf">"Unrecogniced VR code: {0}H - treat as UN"</span>,Integer.toHexString(code));<span style="color:#abb2bf"><em>//return UN;</em></span>LOG.warn(<span style="color:#abb2bf">"zssure:to solve non-standard VR,Unrecogniced VR code: {0}H - treat as UN"</span>,Integer.toHexString(code));<span style="color:#abb2bf">return</span> <span style="color:#abb2bf">null</span>;<span style="color:#abb2bf"><em>//zssure:to solve non-standard VR</em></span>}
</code></span>

2. 正确读取Non-standard VR的VL：

<span style="color:#000000"><code class="language-Java"><span style="color:#abb2bf"><em>//DicomInputStream.java Line 386</em></span><span style="color:#abb2bf">public</span> <span style="color:#abb2bf">int</span> <span style="color:#abb2bf">readHeader</span>() <span style="color:#abb2bf">throws</span> IOException {<span style="color:#abb2bf">byte</span>[] buf = buffer;tagPos = pos; readFully(buf, <span style="color:#abb2bf">0</span>, <span style="color:#abb2bf">8</span>);<span style="color:#abb2bf">switch</span>(tag = ByteUtils.bytesToTag(buf, <span style="color:#abb2bf">0</span>, bigEndian)) {<span style="color:#abb2bf">case</span> Tag.Item:<span style="color:#abb2bf">case</span> Tag.ItemDelimitationItem:<span style="color:#abb2bf">case</span> Tag.SequenceDelimitationItem:vr = <span style="color:#abb2bf">null</span>;<span style="color:#abb2bf">break</span>;<span style="color:#abb2bf">default</span>:<span style="color:#abb2bf">if</span> (explicitVR) {vr = VR.valueOf(ByteUtils.bytesToVR(buf, <span style="color:#abb2bf">4</span>));<span style="color:#abb2bf"><em>//zssure:to solve non-standard VR</em></span><span style="color:#abb2bf"><em>//referred:dcmtk/dcitem.cc/readTagAndLength,Line 970</em></span><span style="color:#abb2bf">if</span>(vr == <span style="color:#abb2bf">null</span>){length = ByteUtils.bytesToUShort(buf, <span style="color:#abb2bf">6</span>, bigEndian);<span style="color:#abb2bf">return</span> tag;                 }<span style="color:#abb2bf"><em>//zssure:end</em></span><span style="color:#abb2bf">if</span> (vr.headerLength() == <span style="color:#abb2bf">8</span>) {length = ByteUtils.bytesToUShort(buf, <span style="color:#abb2bf">6</span>, bigEndian);<span style="color:#abb2bf">return</span> tag;}readFully(buf, <span style="color:#abb2bf">4</span>, <span style="color:#abb2bf">4</span>);} <span style="color:#abb2bf">else</span> {vr = VR.UN;}}length = ByteUtils.bytesToInt(buf, <span style="color:#abb2bf">4</span>, bigEndian);<span style="color:#abb2bf">return</span> tag;}</code></span>