Analysis of X-RAY for pediatry for pathology detection of chest, skeleton & measurements of spine and bones and bone age
TechCare Kids FOR MEASUREMENTS is indicated for patients <18 years old only for the following anatomical regions and views: Left Hand: frontal view Shoulder: glenoid view also known as a ‘true AP’ or a 'Grashey view'Spine: full spine frontal and lateral viewsPelvis: frontal view, weight-bearingHip: Lequesne false-profile view, weight-bearing Leg: frontal and lateral Long-Leg Radiographs, weight-bearing Foot: frontal and lateral views, weight-bearingTechCare Kids is indicated for ages 0 to 18.TechCare Kids is trained to quantify the following measurements:Hand (Left hand):Bone-age Assessment as per the Greulich and Pyle Reference Atlas Shoulder: Critical shoulder Angle (CSA)Spine: Cobb angle, for full-spine and short-spine examinations Coronal Vertical Axis (CVA), Frontal Pelvic Asymmetry (PFA),Thoracic Kyphosis (T1/T12)Lumbar Lordosis (L1/S1), Sagittal Vertical AxisT9 tilt angle or T9 spinopelvic inclination,Pelvic incidence, CAM-HA-offset (offset distance between the center of the acoustic meati (CAM) and hip axis (HA)), also known as Gravity Line, Sacral slope, Pelvic tilt.Pelvis:Lateral center-edge angle (VCE), Acetabular roof angle or Tönnis angle (HTE), Cervical-diaphyseal angle (also known as CC’D angle), Pelvic Frontal Asymmetry (PFA),Femoral offset.Hip: VCA anterior coverage angleLeg (Full-Leg Radiographs): Hip-Knee-Ankle angle (HKA), Femoral neck-shaft angle (NSA) (also known as CC’D angle),Femoral offset, Lengths of Femur, Tibia and total leg, Pelvic tilt Foot:M1-P1 angle (Hallux-valgus angle),P1-P2 angle,M1-M2 angle (First-second Intermetatarsal Angle),M1-M5 angle. Djian-Annonier angle, Calcaneal inclination, First metatarsal pitch angle, Talo-first metatarsal angle (or Meary-Tomeno angle)Published accuracy:Our self-reported accuracy for EACH measure is given in our IFU (appendix A).TechCare Kids FOR PATHOLOGY DETECTION is indicated only for radiographies of the following anatomic regions: Chest (including ribs), clavicle, shoulder, arm, elbow, forearm, wrist, hand, fingers, hip, pelvis, femur, knee, leg (including tibia and fibula), ankle and foot.Depending on the anatomic region and exam view, Techcare Kids is trained to detect the following pathologies:Bone fractureJoint dislocationElbow joint effusionPleural effusion (frontal & profile thorax examination)Pulmonary opacity (frontal thorax examination) Pulmonary nodule (frontal thorax examination)Pneumothorax (frontal thorax examination)
DICOM CX/DX
DICOM GSPS, DICOM SC, DICOM SR
Pathology detection: On 478 pediatric cases: AUC: 0.954 Sentivity: 93.6% / Specificity: 84.6%NPV: 95.9% (1)
Fracture detection: Accuracy 90.1%, Sensitivity 90.2%, Specificity 92.5% (2). Better homogeneity in body subparts and age subgroups compared to other solutions.
Nodule detection: AUC: 0.86, Sensitivity: 50%, Specificity 99%(2)Better performance than radiologist (3)
Chest performance: Airspace disease AUC 0.85, Sensitivity 91%, Specificity 71%, NPV 97% Pneumothorax AUC 0.97, Sensitivity 73%, Specificity 100%, NPV 99%Pleural Effusion AUC 0.94, Sensitivity 78%, Specificity 97%, NPV 95% (4)
For fracture detection: Milvue has been compared to other vendors in external study in university hospital and had the best accuracy and best findings (homogeneity) across all parts of the body and ages sub-groups (Accuracy 90.1%, Sensitivity 90.2%, Specificity 92.5%) (4). This is this only published, external comparative study on IA solutions for Fracture detection in Europe, published in October 2023.
Very high performance for pediatric cases for fracture detection (NPV=95.6%)(5)+20% improvement in sensitivity(2). Milvue is certified from 0 years old. NHS NICE EVA project on Fracture detection recommended the use of Techcare Alert (=TechCare kids) for all population without any age limit where others companies in the market are recommanded above 2 y-oMESURES No restriction in pediatry, certified from 0 years old. Ability to perform measurements even with the presence of foreign bodies (hip prosthesis), post surgical material.Published accuracy (Zerouali et al.), with high precision (Mean Average Error <4° for most Spino-pelvic measurements) and reproducibility. (2)(1)
CASE STUDY for measurements in a private radiology facility: 30k exams per year for trauma, 3k exams per year for measurements (ex full legs, full body…)Detection allows to relief nightshifts and weekends measures, when it takes 15-20mn per radiographer, with Milvue it takes <5mn for control. Before implementing Milvue, two radiographers were assigned to the X-ray machine for handling measurements. With Milvue, one radiographer can manage the entire process. As a result, the second radiographer has been reassigned to other X-Ray Room, thanks to Milvue's automated measurement system. Impact on activity +80%. Before Milvue: 10 exams / 4hWith Milvue: 18 exams / 4h. Productivity gain 1 FTE reallocated to other X-r ray room
CASE STUDY FOR PATHOLOGY DETECTION IN DENMARK Milvue Suite is used in Emergency department of Denmark’s Northern Region through our partner Human BytesIt improves patients’ pathway, reduce waiting time of patient by 1 hour and reduction of complaints*Medical guidelines in the Northern Region have even been changed with the use of Milvue SuiteKey takeaways: Medical personnel in Emergency departments (Physicians, treatment nurses) can decide on a treatment following the output of the IA (ex: further investigation with CT)Patients without fractures can be sent home immediately after their X-RayThe Milvue Suite reduces waiting time and workloadWithin 48h the radiography is reviewed by a radiologist to confirm the diagnosis (5)
(1) Parpaleix A, Parsy C, Cordari M, Mejdoubi M. Assessment of a combined musculoskeletal and chest deep learning-based detection solution in an emergency setting. Eur J Radiol Open. 2023 Mar 10;10:100482. doi: 10.1016/j.ejro.2023.100482. PMID: 36941993; PMCID: PMC10023863
(2) Bousson et al, Academic Radiology, 2023
(3) Comparison of Commercial AI Software Performance for Radiograph Lung Nodule Detection and Bone Age Prediction. Kicky G. van Leeuwen, 2024
(4) Lind Plesner L , C. Müller F, W. Brejnebøl M, C. Laustrup L, Rasmussen F, W. Nielsen O, Boesen M, Brun Andersen M.Publication : September 2023
(5) https://rn.dk/da/Service/Nyhedsbase-RN/2024/02/Kunstig-intelligens-reducerer-baade-ventetid-og-antallet-af-klagesager
