Files
rustytorch/demos/rtx-tumorboard-demo/src/ct_analyzer.rs
T
2026-03-04 00:08:42 +00:00

359 lines
11 KiB
Rust

//! CT scan analysis using 3D CNN.
//!
//! Analyzes CT scans to detect lesions, measure dimensions, and characterize findings.
use tumorboard_shared::{
CTFindings, ImagingStudy, Lesion, LesionLocation, LesionType, OrganMeasurement, RegionHU,
};
/// Error type for CT analysis.
#[derive(Debug, thiserror::Error)]
pub enum CTAnalysisError {
/// No scans provided.
#[error("No CT scans provided")]
NoScans,
/// Invalid scan data.
#[error("Invalid scan data: {0}")]
InvalidData(String),
}
/// CT scan analyzer using simulated 3D CNN.
#[derive(Debug)]
pub struct CTAnalyzer {
/// Detection threshold.
detection_threshold: f32,
/// Minimum lesion size (mm).
min_lesion_size: f32,
}
impl Default for CTAnalyzer {
fn default() -> Self {
Self::new()
}
}
impl CTAnalyzer {
/// Create a new CT analyzer.
#[must_use]
pub fn new() -> Self {
Self {
detection_threshold: 0.5,
min_lesion_size: 3.0,
}
}
/// Create with custom thresholds.
#[must_use]
pub fn with_thresholds(detection_threshold: f32, min_lesion_size: f32) -> Self {
Self {
detection_threshold,
min_lesion_size,
}
}
/// Analyze CT studies.
///
/// # Errors
///
/// Returns error if no studies are provided or data is invalid.
pub fn analyze(&self, studies: &[&ImagingStudy]) -> Result<CTFindings, CTAnalysisError> {
if studies.is_empty() {
return Err(CTAnalysisError::NoScans);
}
// Detect lesions
let lesions = self.detect_lesions(studies);
// Get organ measurements
let organ_measurements = self.measure_organs(studies);
// Analyze Hounsfield units
let hu_analysis = self.analyze_hu(studies);
// Generate assessment
let assessment = self.generate_assessment(&lesions);
// Calculate confidence
let confidence = self.calculate_confidence(studies, &lesions);
Ok(CTFindings {
lesions,
organ_measurements,
hu_analysis,
assessment,
confidence,
})
}
/// Detect lesions in the CT studies.
fn detect_lesions(&self, studies: &[&ImagingStudy]) -> Vec<Lesion> {
let mut lesions = Vec::new();
// Simulate lesion detection based on study properties
// In real implementation, this would use a 3D CNN
for study in studies {
for image in &study.images {
// Check body region for lung lesion simulation
if matches!(image.body_region, tumorboard_shared::BodyRegion::Chest) {
// Primary lung lesion
let primary_lesion = Lesion {
id: "L001".to_string(),
location: LesionLocation {
organ: "Lung".to_string(),
region: "Right upper lobe".to_string(),
center: [45.0, 120.0, 85.0],
},
size: [32.0, 28.0, 25.0],
volume: 11_700.0,
lesion_type: LesionType::Nodule,
malignancy_probability: 0.85,
confidence: 0.92,
};
lesions.push(primary_lesion);
// Possible satellite nodule (higher res imaging)
if image.spacing[2] <= 2.0 {
let satellite = Lesion {
id: "L002".to_string(),
location: LesionLocation {
organ: "Lung".to_string(),
region: "Right upper lobe, satellite".to_string(),
center: [52.0, 115.0, 82.0],
},
size: [8.0, 7.0, 6.0],
volume: 175.0,
lesion_type: LesionType::Nodule,
malignancy_probability: 0.65,
confidence: 0.78,
};
lesions.push(satellite);
}
}
}
}
// Filter by detection threshold
lesions
.into_iter()
.filter(|l| l.confidence >= self.detection_threshold)
.filter(|l| l.size[0] >= self.min_lesion_size)
.collect()
}
/// Measure organs.
fn measure_organs(&self, studies: &[&ImagingStudy]) -> Vec<OrganMeasurement> {
let mut measurements = Vec::new();
for study in studies {
for image in &study.images {
if matches!(image.body_region, tumorboard_shared::BodyRegion::Chest) {
// Heart measurements
measurements.push(OrganMeasurement {
organ: "Heart".to_string(),
measurement_type: "Cardiothoracic ratio".to_string(),
value: 0.48,
unit: "ratio".to_string(),
normal_range: Some((0.40, 0.50)),
is_abnormal: false,
});
// Aorta
measurements.push(OrganMeasurement {
organ: "Aorta".to_string(),
measurement_type: "Ascending diameter".to_string(),
value: 35.0,
unit: "mm".to_string(),
normal_range: Some((20.0, 37.0)),
is_abnormal: false,
});
}
}
}
measurements
}
/// Analyze Hounsfield units.
fn analyze_hu(&self, _studies: &[&ImagingStudy]) -> Vec<RegionHU> {
vec![
RegionHU {
region: "Lesion L001".to_string(),
mean: 35.0,
std_dev: 12.0,
interpretation: "Soft tissue density, consistent with solid mass".to_string(),
},
RegionHU {
region: "Normal lung".to_string(),
mean: -850.0,
std_dev: 50.0,
interpretation: "Normal aerated lung parenchyma".to_string(),
},
]
}
/// Calculate overall confidence.
fn calculate_confidence(&self, studies: &[&ImagingStudy], lesions: &[Lesion]) -> f32 {
let mut confidence: f32 = 0.7;
// Better resolution = higher confidence
for study in studies {
for image in &study.images {
if image.spacing[2] <= 1.0 {
confidence += 0.15;
} else if image.spacing[2] <= 2.0 {
confidence += 0.1;
}
}
}
// More lesions detected = more confident in analysis
if !lesions.is_empty() {
confidence += 0.05;
}
confidence.min(0.98)
}
/// Generate overall assessment.
fn generate_assessment(&self, lesions: &[Lesion]) -> String {
if lesions.is_empty() {
return "No significant lesions identified.".to_string();
}
let primary = &lesions[0];
let mut assessment = format!(
"Suspicious {:?} in {} measuring {:.1} x {:.1} x {:.1} mm.",
primary.lesion_type,
primary.location.region,
primary.size[0],
primary.size[1],
primary.size[2]
);
if primary.malignancy_probability > 0.7 {
assessment.push_str(" Highly suspicious for malignancy.");
} else if primary.malignancy_probability > 0.4 {
assessment.push_str(" Intermediate suspicion for malignancy.");
}
if lesions.len() > 1 {
assessment.push_str(&format!(
" {} additional nodule(s) identified.",
lesions.len() - 1
));
}
assessment
}
}
#[cfg(test)]
mod tests {
use super::*;
use tumorboard_shared::{BodyRegion, ImageDimensions, ImageMetadata, ImagingModality};
fn create_test_study() -> ImagingStudy {
ImagingStudy {
study_id: "CT001".to_string(),
modality: ImagingModality::CT,
images: vec![ImageMetadata {
id: "CT001-001".to_string(),
modality: ImagingModality::CT,
acquisition_date: "2024-01-15".to_string(),
dimensions: ImageDimensions {
width: 512,
height: 512,
depth: 300,
},
spacing: [0.7, 0.7, 1.25],
body_region: BodyRegion::Chest,
series_description: Some("Chest CT".to_string()),
}],
description: Some("Chest CT with contrast".to_string()),
}
}
#[test]
fn test_ct_analyzer_creation() {
let analyzer = CTAnalyzer::new();
assert!(analyzer.detection_threshold > 0.0);
}
#[test]
fn test_analyze_single_study() {
let analyzer = CTAnalyzer::new();
let study = create_test_study();
let result = analyzer.analyze(&[&study]);
assert!(result.is_ok());
let findings = result.unwrap();
assert!(!findings.lesions.is_empty());
assert!(findings.confidence > 0.0);
}
#[test]
fn test_no_studies_error() {
let analyzer = CTAnalyzer::new();
let result = analyzer.analyze(&[]);
assert!(result.is_err());
}
#[test]
fn test_lesion_detection() {
let analyzer = CTAnalyzer::new();
let study = create_test_study();
let findings = analyzer.analyze(&[&study]).unwrap();
// Should detect the primary lung lesion
assert!(!findings.lesions.is_empty());
let primary = &findings.lesions[0];
assert!(primary.location.region.contains("upper lobe"));
assert!(primary.malignancy_probability > 0.5);
}
#[test]
fn test_organ_measurements() {
let analyzer = CTAnalyzer::new();
let study = create_test_study();
let findings = analyzer.analyze(&[&study]).unwrap();
assert!(!findings.organ_measurements.is_empty());
}
#[test]
fn test_hu_analysis() {
let analyzer = CTAnalyzer::new();
let study = create_test_study();
let findings = analyzer.analyze(&[&study]).unwrap();
assert!(!findings.hu_analysis.is_empty());
}
#[test]
fn test_custom_thresholds() {
let analyzer = CTAnalyzer::with_thresholds(0.9, 10.0);
let study = create_test_study();
let findings = analyzer.analyze(&[&study]).unwrap();
// Higher threshold may filter out some lesions
// The primary lesion has confidence 0.92, so it should still be detected
assert!(!findings.lesions.is_empty());
}
#[test]
fn test_assessment_generation() {
let analyzer = CTAnalyzer::new();
let study = create_test_study();
let findings = analyzer.analyze(&[&study]).unwrap();
assert!(!findings.assessment.is_empty());
assert!(findings.assessment.contains("Suspicious"));
}
}