Default ValueReadStrategy about dicom-rs HOT 6 CLOSED

Please use to_multi_f64 instead. float64_slice is a non-converting method, and would only work for FD-type or OD-type elements.

Thank you, it seems to work!

from dicom-rs.

Thank you for reaching out. Note that changing the parser reading strategy is not necessary for that use case. It only means that a conversion needs to take place after the parsing, which is the standard way of working with DICOM values at the moment. To treat a DS value as a floating point number:

// example
let val = DicomValue::new(dicom_value!(Strs, ["1300", "664 "]));

// converts strings automatically
let ww = val.to_float64()?;
// for multiple window levels, returns a vector
let ww = val.to_multi_float64()?;

The documentation on DicomValue has the full list of methods available. The ones named string, uint16 (and so on) only check their internal representation without conversion, whereas methods named to_* will try to convert the internal data if necessary. Note also that retrieving a simple enum value describing the value's type can be done through the method value_type.

Changing the value reading strategy requires creating the data set reader with the function DataSetReader.new_with_dictionary, then passing it manually to InMemDicomObject.from_element_source. This is not even sure to work well. The main reason for this is that any non-compliant value, such as a date or time value in another format, could prevent the parser from reading the rest of the data set. Context: #9 #10

from dicom-rs.

Thank you @Enet4 for the detailed explanation. I got the idea of conversion on retrieval. Well... dates/times can be treated this way: storing strings but retrieving dates/times. However please explain why you did the same for floating-point values in contrast to integer values stored natively? Are you afraid of meeting a non-standard floating-point value, for example, with a decimal comma instead of a dot? Thank you for following up on this conversation!

from dicom-rs.

However please explain why you did the same for floating-point values in contrast to integer values stored natively? Are you afraid of meeting a non-standard floating-point value, for example, with a decimal comma instead of a dot?

That would be a good point, and a justification to continue with this strategy, but the reasoning of this being the default is primarily to preserve the original representation: any textual value stays textual, thus preventing failures during the parsing phase and other oddities (such as decimal numbers losing precision when turned to a binary type).

This does not apply uniquely to dates/times and floating point values. If the VR is IS (e.g. for Instance Number), it will still be kept as a string. This is fine, as conversion is trivially done at a later stage through to_int or any of the other conversion methods.

from dicom-rs.

OK @Enet4, thanks for clarifying the architecture.

Now I would like to share a real code converting DICOM elements into JSON values (some of them at least):

fn value_to_json<I, P>(element: &DataElement<I, P>) -> Option<JsonValue> {
    if element.header().len.get().unwrap() == 0 {
        return None;
    }
    println!("VR: {}", element.vr());
    let value = element.value();
    if value.primitive().is_some() {
        return match value.multiplicity() {
            0 => None,
            1 => single_value_to_json(value),
            _ => multi_value_to_json(value),
        };
    }
    println!("Skip {} as non-primitive element", element.header().tag);
    None
}

fn single_value_to_json<I, P>(value: &DicomValue<I, P>) -> Option<JsonValue> {
    if let Ok(value) = value.float64() {
        println!("float64");
        return Some(JsonValue::from(value));
    }
    if let Ok(value) = value.uint16() {
        println!("uint16");
        return Some(JsonValue::from(value));
    }
    if let Ok(value) = value.string() {
        println!("string");
        return Some(JsonValue::from(value));
    }
    None
}

fn multi_value_to_json<I, P>(value: &DicomValue<I, P>) -> Option<JsonValue> {
    if let Ok(value) = value.uint16_slice() {
        println!("uint16[]");
        return Some(JsonValue::from(value));
    }
    if let Ok(value) = value.float64_slice() {
        println!("float64[]");
        return Some(JsonValue::from(value));
    }
    if let Ok(value) = value.strings() {
        println!("string[]");
        return Some(JsonValue::from(value));
    }
    None
}

It traces the following for a file:

VR: UI
string
VR: UI
string
VR: DA
string
VR: TM
string
VR: CS
string
VR: LO
string[]
VR: CS
string
VR: UI
string
VR: UI
string
VR: US
uint16
VR: CS
string
VR: US
uint16
VR: US
uint16
VR: DS
string[]
VR: US
uint16
VR: US
uint16
VR: US
uint16
VR: US
uint16
VR: LT
string

Please note that value.float64_slice seems to return None for a multi-value DS-type element. I see a similar issue for single-value elements: value.float64 seems to return None for them. Could you explain why?

from dicom-rs.

Please use to_multi_f64 instead. float64_slice is a non-converting method, and would only work for FD-type or OD-type elements.

from dicom-rs.