The initial search identified 2,692 studies, from which 1,614 duplicates were removed with a further 813 removed after title and abstract screening. An additional 176 were removed following full text screening, which resulted in 78 studies from 81 publications that met the inclusion criteria (Additional file 2 and 3).
Trends in publication of MRI studies: 1992 to 2020
Frequency of publication of MRI studies has increased steadily since 1992, growing from one study (43) to 12 in 2020 (44-55) (Figure 1A). Across the included studies, 17 countries were represented: Australia (n=16), Japan (n= 11), Germany (n= 10), USA (n= 10), United Kingdom (n= 7), Switzerland (5), Finland (n= 4), France (n= 3), Netherlands (n= 3), Spain (n= 3), China (n=2), Turkey (n= 2), Canada (n=1), New Zealand (n= 1), Norway (n= 1), Poland (n= 1). A range of study designs were used including nine randomised controlled trials, 33 prospective cohort, 10 retrospective cohort, 15 case-control and 10 case series study designs (Figure 1B).
Patient and non-patient populations
Twenty-three studies across 25 publications investigated hip related musculoskeletal pain (e.g., hip osteoarthritis, lateral hip pain and intra-articular hip joint pathologies) (Figure 1C). Three studies examined non-hip related musculoskeletal pain which included low back pain (56, 57) and patellofemoral joint osteoarthritis (58). Twenty-nine studies, across 33 publications, used healthy comparison groups and 26 studies explored one of three surgical presentations (i.e., total hip arthroplasty, hip arthroscopy and surgical correction for hip dysplasia) (Figure 1C). Gluteus medius was the most frequently assessed lateral hip muscle (Figure 1D). Fifty-four studies measured muscle size and 40 studies investigated fatty infiltration (Table 1).
Measurement of muscle size and quality
Thirty-six studies reported the profession of the individual(s) interpreting MRIs and calculating size and fatty infiltration measures. The most frequently cited professionals were radiologists (31 studies) with 15 studies reporting radiologists with further training in musculoskeletal presentations. Other health professionals included orthopaedic surgeons and physiotherapists. Ten studies (44, 57, 59-66) reported years of experience for those who interpreted the MRIs, which ranged from 1 year to 28 years.
For volume measures, ICC scores reflected good to excellent reliability, with data ranging from 0.87 to 0.99 for intra-reliability and 0.80 to 0.99 for inter-reliability. For CSA, reliability was moderate to excellent with scores ranging from 0.70 to 1.00 for intra-rater reliability and 0.70 to 0.99 for inter-rater reliability. Fatty infiltration ICC values also indicated moderate to excellent reliability with scores ranging from 0.70 to 0.99 for intra-rater reliability and 0.70 to 0.98 for inter-rater reliability. The kappa coefficient scores indicated fair to almost perfect agreement between and among studies (intra-rater reliability k range 0.463 - 0.93; inter-rater reliability k range, 0.51- 0.91) (Table 1). No study reported scan to rescan reliability.
MRI parameters
The MRI parameters of all studies are summarised in Table 1. Two MRI field strengths were reported, 1.5 Tesla and 3 Tesla. A wide range of MRI sequences were used across the studies, with many incorporating several sequence types, both T1- and T2-weighted, with and without fat suppression. Slice thickness ranged from 0.5mm to 15mm, with 16 studies (20.3%) not reporting slice thickness. Acquisition time ranged from 2hrs 32mins (67) to 1min 29s (68).
All studies that reported patient positioning specified a supine position with legs extended and hips in neutral, except three studies (45, 69, 70) that used pillows under the knees for comfort, and two studies (44, 46) placing the hips into internal rotation.
Table 1 MRI parameters for individual studies
Muscle size measures
Lateral hip muscle volume was measured in 31 studies and CSA was measured in 24 studies, (Table 2 and 3). For volume measures, manual segmentation techniques were most frequently used (77.4%) compared to automated. For CSA, all studies used manual segmenting techniques.
Volume measurement outcomes
Whole muscle volume was calculated for 28 studies (90.3%), while two (26, 52) measured partial muscle volume. To calculate volume, all studies incorporated sums of CSA estimates. Seventeen (54.8%) studies also incorporated slice thickness and five (16.1%) normalised calculations to either individual height or mass (Table 2).
Table 2 Volume measurement outcomes for individual studies
Cross-sectional area measurement outcomes and axial anatomical slice location
Five studies calculated CSA from multiple slices either by using the mean derived from several consecutive slices or assessing CSA at two predetermined locations (Table 3). Single axial slices were chosen at a pre-determined anatomical locations for all other studies except for two studies (71, 72), which measured at the single slice with the greatest CSA for the individual muscle.
Table 3 Cross sectional area measurement outcomes for individual studies
Seven anatomical levels were identified as locations where CSA can be measured for the lateral hip muscles (Figures 2, 3 & 4). These include i) anterior superior iliac spine (ASIS) (73-75) ii) half way between the iliac crest and the superior tip of the greater trochanter (59) iii) anterior inferior iliac spine (AIIS) (73) iv) upper border of the acetabulum (46, 76, 77) v) superior tip of the greater trochanter (45, 62, 69, 78-81) vi) lower border of the acetabulum (25, 76, 77) and vii) lesser trochanter (57)(75).
Fig. 2 3-D representation of anatomical levels for each lateral hip muscle
When comparing MRI images to E-12 anatomical plastinates (Fig. 3 & 4), the E-12 anatomical plastinates provide better visualisation of muscle borders. At levels AIIS and the upper border of the acetabulum, the muscle borders between gluteus medius and piriformis are better visualised on the E-12 anatomical plastinates with detail of individual muscle fibre directions demarcating the individual muscles (Fig. 5). For levels at superior tip of greater trochanter and below, the TFL border is better visualised on the E-12 anatomical plastinates against neighbouring muscle borders including the gluteus medius and rectus femoris.
Fig. 3 Axial DIXON sequence MRI and E-12 anatomical plastinate comparison at anatomical levels for cross sectional area measurement above the hip joint. A At the level of anterior superior iliac spine B Halfway between the iliac crest and the superior tip of the greater trochanter C Anterior inferior iliac spine; square dotted box surrounds enlarged morphological region of interest (fig. 4); 1- gluteus minimus; 2- gluteus medius; 3- gluteus maximus; 4- TFL; 5- ilium; 6- iliacus; 7- psoas major; 8- rectus abdominis
Fig. 4 Axial DIXON sequence MRI and E-12 anatomical plastinate comparison at anatomical levels for cross sectional area measurement at and below hip joint. A upper border of the acetabulum B superior tip of the greater trochanter C lower border of the acetabulum D lesser trochanter; 1- gluteus minimus; 2- gluteus medius; 3- gluteus maximus; 4- TFL; 5- acetabulum; 6- Iliacus; 9- Acetabulum; 10- Piriformis; 11- Iliopsoas; 12- Sartorius; 13-Rectus femoris; 14- Femoral head; 15- Greater trochanter; 16- lesser trochanter; 17- Vastus lateralis; 18- Pectineus; 19: Adductor brevis; 20- Adductor magnus; 21- Quadratus femoris
Fig. 5 Enlarged region of interest at the level of anterior inferior iliac spine. A Axial DIXON sequence MRI B E-12 anatomical plastinate C Schematic illustration; round circle indicates feature of interest; Red line- gluteus minimus; Green line- gluteus medius; Dashed red line- partition between gluteus medius and piriformis; Dashed grey line- partition between gluteus maximus with both gluteus medius and piriformis; Red circle- highlights angles between partitions to help identify separation between piriformis and gluteus medius
Some same slice locations were described in multiple ways as these levels contained multiple identifying features. For example the slice location at the level of the tip of the greater trochanter (level vi) is consistent with the level described as the centre of the femoral head (62, 79, 80), and the level where the femoral head has the greatest CSA (45), depending on slice thickness. Other slice locations were at a pre-set distance from an anatomical feature including 20mm distal to the proximal aspect of the femoral head (82) for gluteus maximus and 15mm from the superior margin of the acetabulum (67) for gluteus medius and minimus.
Intramuscular fatty infiltration measurement outcomes and axial anatomical slice location
Forty studies measured intra-muscular fatty infiltration (Table 4). Qualitative measures of fatty infiltrate were used by 30 studies with the Goutellier classification being the most frequently used. Quantification methods, using a ratio of pixel intensity from fat and water images were used by 10 studies. This technique has become more utilised over recent years.
Gluteus medius and/or gluteus minimus were further divided into compartments in 11 studies. Gluteus medius was divided into three equal compartments (anterior, middle and posterior) by nine studies and two equal compartments (anterior, posterior) by one study. Similarly, gluteus minimus was divided into three equal compartments (anterior, middle and posterior) by seven studies and into two equal parts (anterior and posterior) by two studies. The TFL and gluteus maximus were not divided into compartments for intramuscular fatty infiltration measurement.
Table 4 Fatty infiltration measurement outcomes for individual studies
Six anatomical levels were identified as locations for fatty infiltration measurement of the lateral hip muscles (Figure 6). Two levels were identified for TFL, four levels were identified for gluteus maximus, gluteus medius and gluteus minimus muscles. Four studies (53, 83-85) described quantitative measures of fatty infiltration for whole muscle.
Tensor fascia latae
The two anatomical levels for TFL fatty infiltration assessment included the superior tip of the greater trochanter (79, 81) and the lesser trochanter (67, 74, 75). The level at the greater trochanter was consistent with other anatomical features including the centre of the femoral head (79) and the fovea capitis (19, 21, 22). The ischial tuberosity was described in one study (86) and can span multiple slices. The greatest axial CSA was described in one study (87).
Gluteus maximus
The four levels for gluteus maximus fatty infiltration assessment are i) the distance at one third the distance from the iliac crest to the superior tip of the greater trochanter(19) ii) greater sciatic foramen (superior most part)(19, 21, 22, 42) iii) two thirds the distance from the iliac crest to the superior tip of the greater trochanter(19) iv) the superior tip of the greater trochanter(19, 88). The level where the femoral head has a round configuration(66) and where it has the greatest circumference(19) was deemed similar to the level at the greater trochanter.
Gluteus medius and minimus
Gluteus medius and gluteus minimus were frequently assessed individually at the same level within a study. The four levels for gluteus medius and gluteus minimus fatty infiltration assessment are i) the distance at one third the distance from the iliac crest to the superior tip of the greater trochanter (19, 59, 61, 79, 87) ii) anterior superior iliac spine (74, 75) iii) greater sciatic foramen (superior most part) (19, 21, 22, 42) and iv) two thirds the distance from the iliac crest to the superior tip of the greater trochanter(19, 59, 61, 79, 87, 89-92).
Other levels described included pre-determined distances from anatomical features and included 15 mm superior to the upper margin of the acetabulum(67), three and six slices proximal to greater trochanter with slice thickness set at 6mm (93), 30 mm proximal to greater trochanter (94). Descriptions of levels that could span multiple axial slices included the level of the acetabulum(67, 88) and the ipsilateral sacroiliac joint (95).
Fig. 6 Anatomical levels of interest for fatty infiltration measurement outcomes
Machine learning
Overall machine learning was incorporated in 16 (20.3%) of the studies. For size measures, eight (25.8%) studies reporting volume either used automatic or semi-automatic tracing methods while no study reporting CSA incorporated machine learning. For fatty infiltration, 10 (25.0%) studies used machine learning to identify and quantify water and fat value pixels within regions of interest.