Imaging biomarkers of leukaemic choroidopathy

To longitudinally investigate choroidal and choriocapillaris perfusion metrics and the number of choroidal hyperreflective foci (HRF) in patients with acute leukaemia (AL) before and after disease remission and to correlate these metrics with systemic parameters during active disease.

An in vivo characterization of choroidal changes in AL is currently lacking. This study aimed to fill this knowledge gap. Firstly, we compared the choroidal perfusion metrics between AL patients and healthy controls and explored their correlations with systemic haematological parameters. Then, we investigated the choroidal changes before and after disease remission. With the hypothesis that HRF may represent an indirect sign of leukaemic choroidopathy, we also included the assessment of HRF in our analyses.

| M ET HOD S
This was a prospective, longitudinal study on AL patients treated at the Haematology and Bone Marrow Transplant Unit of San Raffaele Scientific Institute, Milano (Italy) from March 2021 to December 2021. We conducted the study under the tenets of the Declaration of Helsinki, and we obtained written informed consent from each patient. We included patients affected by myeloid or lymphoblastic AL, diagnosed based on morphological, cytogenetical, immune-phenotypical and molecular studies on the bone marrow aspirate. We excluded eyes with optic media opacities, retinal or optic disc tumour infiltration, or other concurrent ocular diseases.
All patients underwent systemic chemotherapy at the Haematology and Bone Marrow Transplant Unit. Patients underwent bone marrow biopsy 1 month after consolidation therapy to confirm AL remission, defined by a total or partial reduction in the bone marrow blasts' count below the threshold of 5%. Patients with AL remission were re-examined the same days they were hospitalized for bone marrow biopsy.
We recruited an age-and gender-matched healthy subject as a control for each included eye.
We exported the structural SD-OCT scans and the OCTA slabs as JPEGs. We analysed all the images with the free software ImageJ (National Institutes of Health, Bethesda, MD, USA).

| Choroidal measurements
On SD-OCT scans, we manually measured the subfoveal choroidal thickness (CT) as the distance between Bruch's membrane and the sclero-choroidal junction under the foveal depression. We measured the total (TCA), the luminal (LCA) and the stromal choroidal area (SCA) in a 1500μm wide region centred on the fovea. We selected the TCA as a 1500μm polygon having Bruch's membrane as the superior boundary and the sclero-choroidal junction as the inferior boundary. We selected the LCA using the colour threshold tool (hue and saturation bars set on 0-255, brightness bars set on 0 and 254) after binarizing each SD-OCT scan with the Niblack autolocal threshold. We calculated the CVI as the ratio between the LCA and the TCA. We derived the SCA as the difference between the TCA and the LCA (Betzler et al., 2022).
We identified HRF as round or diamond-shaped dots with hyperreflective signal on B-scan SD-OCT scans. To ensure a standardized and reproducible method for HRF count, we set a reflectivity threshold, selecting those pixels with a signal brightness of at least 2 standard deviations (SD) above the mean choroidal brightness (Corradetti et al., 2019;Kim & Oh, 2022;Nassisi et al., 2018). As a preliminary step, we corrected the choroidal signal intensity for any potential attenuation by optic media opacity normalizing all SD-OCT images according to vitreous and the retinal pigment epithelium (RPE) mean brightness ( Figure 1a). This adjustment was fundamental to avoid hyper-or hyporeflectivity due to shadowing artefacts. All images were normalized, assuming the vitreous and RPE brightness to be 0 (black) and 255 (white) respectively. We measured the mean vitreous brightness in a 1500 × 1500μm square area upon the fovea through the imageJ histogram tool. We subtracted the obtained value from the whole image intensity using the subtract tool (Process → Math → Subtract). Then, we divided each pixel's brightness by the mean RPE brightness, measured in a 1500μm wide region under the fovea, according to the following formula: RPE mean brightness/white (255) (Borrelli et al., 2017). After normalization of the OCT scan brightness (Figure 1b), we manually segmented the choroidal area with the polygon selection tool, using the RPE as the superior, the sclero-choroidal junction as the inferior, and the scan margins as lateral boundaries ( Figure 1c). Then, we cleared the image outside the selected region of interest.
To highlight HRF, we computed the mean choroidal brightness using the histogram tool and subtracted the mean brightness +2 SD (Figure 1d) (Corradetti et al., 2019;Kim & Oh, 2022;Nassisi et al., 2018). We increased the image contrast with the enhance contrast tool (0.3% of saturated pixels) (Figure 1e), and we binarized the resulting image using the Phalsankar autolocal threshold with a preset radius of 4 pixels ( Figure 1f). We automatically counted the number of choroidal HRF with the Analyse particles function (Corradetti et al., 2019, Kim & Oh, 2022, Nassisi et al., 2018. We tested the consistency of our HRF counting method by computing the inter-rater agreement between two independent raters (MM and AA). We used with intraclass correlation coefficient (ICC) with a two-way random-effects model (Koo & Li, 2016). The ICC was moderate (ICC = 0.72, 95% CI 0.56 to 0.83), and we used the averaged HRF counts between the two readers for subsequent analyses.

| Choriocapillaris measurements
We measured the FD density in a 20μm-thick choriocapillaris slab, automatically generated by the OCTA software (TVue XR AngioVue Optical Coherence Tomography Angiography Software) (Nesper et al., 2021) to be consistent with the previous literature using the same device. To reduce the shadow effect from the overlying large intraretinal vessels, we generated a mask from the superficial retinal capillary plexus through max entropy thresholding ( Figure S1a, b), and we subtracted the mask from the choriocapillaris slab ( Figure S1c). We binarized the obtained image with the Phalsankar autolocal threshold with a preset 4-pixel radius (Figure S1d), and we automatically counted the area with no flow signal with the analyse particle tool. We corrected the FD density (cFD density) by subtracting the applied mask from the entire slab area, according to the formula:

| Statistical analysis
We performed all statistical analyses by setting a threshold for statistical significance at p < 0.05. We summarized continuous variables as their mean ± SD if normally distributed or as their median and interquartile range (IQR) otherwise. We summarized categorical variables as their absolute and relative prevalence.
As per the primary aim, we investigated differences in CT, TCA, LCA, SCA, CVI, cFD density and choroidal HRF number between AL and control eyes using linear mixed models where each quantitative metric was the dependent variable, the group (AL vs. control) was the fixed effect. The patients' identification number was the random effect to account for intrasubject correlations for both eyes' inclusion. We reported regression estimates and their 95% confidence intervals (CI).
As per the secondary aim, we explored correlations between the choroidal metrics, namely CT, TCA, LCA, SCA, CVI, cFD density and choroidal HRF, and laboratory findings (WBC, circulating blasts, haemoglobin, haematocrit, platelets, fibrinogen, D-dimer, INR, and APTT ratio) in AL patients using Pearson's correlations, and we reported correlation coefficients (r) and their 95% CI.
Finally, we investigated the longitudinal changes in choroidal perfusion metrics and choroidal HRF number within patients who reached AL remission using linear mixed models, where both the eye' and the patients' identification numbers were the nested random effects to account for intraeye and intrasubject correlations.

| R E SU LT S
Twenty-six eyes of 14 AL patients (9 males, 64%; mean age 59 ± 12 years) and 26 control eyes of 26 from 13 healthy volunteers were included. Two eyes of two patients were excluded because of retinal co-morbidities. Namely, one eye was excluded due to branch retinal arterial occlusion and the other due to previous retinal detachment treated with trans pars plana vitrectomy. Eleven patients (79%) had a new diagnosis of AL, three patients (11%) were studied at the time of disease relapse. None of the 14 patients had ocular symptoms, and the VA was 20/20 in all eyes. The cohort overlaps with a previous paper from our group investigating retinal circulation in AL (Cicinelli et al., 2022). Clinically visible leukaemic retinopathy was observed in seven eyes of five patients (27%). Specifically, cotton wool spots were found in six eyes, mid-peripheral blot haemorrhages in two eyes and peripapillary splinter haemorrhages in one. None of the 26 eyes had signs of leukaemic retinopathy within the macula, potentially influencing the quality of the images.

| Choroidal and choriocapillaris metrics between AL patients and healthy controls
The TCA (1.65 ± 0.43 mm 2 vs. 1.38 ± 0.41 mm 2 , p = 0.028), the LCA (1.06 ± 0.27 mm 2 vs. 0.91 ± 0.26 mm 2 , p = 0.044) and the SCA (0.59 ± 0.19 mm 2 vs. 0.48 ± 0.17 mm 2 , p = 0.024) were significantly higher in AL eyes than controls. On the contrary, the subfoveal CT, the CVI and the cFD density did not significantly differ between the two groups (Table 1). No significant differences in the choroidal metrics were found between eyes with leukaemic retinopathy and those without retinopathy (all p > 0.05).
The choroidal HRF were more numerous in the patients' group (88 ± 24 vs. 67 ± 15, p = 0.001). Choroidal HRF were predominantly arranged superficially in the choroid, presumably in the Sattler's layer, and clustered in the stromal area around small vessels. They did not show a preferential disposition between the subfoveal and the extrafoveal area ( Figure S2). In the AL group, there was an association between the number of choroidal HRF and the TCA (regression estimate = 24.9 for each TCA mm 2 increase, 95% CI 3.00 to 40.7, p = 0.001) and the SCA (regression estimate = 49.7 for each SCA mm 2 increase, 95% CI 10.9 to 91.4, p = 0.01), but not with the LCA (regression estimate = 23.7 for each LCA mm 2 increase, 95% CI −5.36 to 48.05, p = 0.06). No significant correlation was found in the control group.
A higher cFD density was associated with a higher WBC count (r = 0.45, 95% CI 0.04 to 0.67, p = 0.036). No other correlations were found among the remaining laboratory tests and choroidal metrics (Table S1).
Postremission subfoveal CT, TCA, LCA, SCA, cFD density and choroidal HRF count were similar to the control group values.

| DI SC US SION
This study longitudinally investigated the choroidal metrics in patients with AL by means of SD-OCT and OCTA and their correlations with systemic parameters. To date, it is the first in vivo deep characterization of choroidal involvement in patients with AL. We found patients with AL in their active phase had higher luminal and stromal choroidal areas than healthy subjects and more numerous choroidal HRF. The choriocapillaris was only marginally involved in the active disease, although our data suggest an inverse correlation between choriocapillaris perfusion, leukocytosis and procoagulant state. The choroid is a privileged homing site of leukaemic cells (Dick & Lapidot, 2005). Leonardy et al. (1990) analysed postmortem eyes of patients with leukaemia and reported 93% of them had choroidal infiltration. Allen & Straatsma (1961) examined the eyes of 76 patients who died from leukaemia and reported an 89% rate of choroidal involvement. Although a diffuse choroidal thickening on SD-OCT has been described in patients with leukaemia (Bajenova et al., 2012; Choi, 2020 ;Coupland & Damato, 2008;Takita et al., 2018;Vieira et al., 2015), available imaging studies are mostly anecdotal. None included a quantitative analysis of the choroid and the choriocapillaris.
In our study, we separately investigated the different choroidal components, and we found patients with active AL had supranormal values of SCA, LCA and TCA. While the LCA remained unchanged, the SCA significantly reduced after disease remission, suggesting the largest changes in active AL occurred in the choroidal stroma. Postmortem studies of AL patients reported mottled thickening of the choroid and sheets of neoplastic cells in the innermost stromal interstices (Allen & Straatsma, 1961), more abundant at the posterior pole (Leonardy et al., 1990). In accordance with histopathology descriptions, we hypothesize that the reversible increase in the SCA values could be related to neoplastic infiltration. Other plausible alternatives, however, include an increased inflow of fluids into the choroidal interstices due to blood flow congestion or a generalized thickening due to a tumour-induced inflammatory environment (Mitamura et al., 2022). We observed more choroidal HRF in patients with AL than controls, which considerably reduced after disease remission. We speculate choroidal HRF represent a biomarker of leukaemic choroidopathy. Intraretinal HRF have been characterized in several diseases, such as diabetic macular oedema (Saurabh et al., 2021), angioid streaks (Romano et al., 2019) and age-related macular degeneration (Nassisi et al., 2018). Conversely, choroidal HRF have been seldom explored and have been interpreted as lipofuscin deposits (Piri et al., 2015;Saurabh et al., 2021), macrophages (Kim & Oh, 2022;Romano et al., 2020), degenerated RPE cells (Romano et al., 2020) or melanocytes (Kim & Oh, 2022). Hyperreflectivity on OCT is an optical property of different molecules, and we cannot univocally conclude on HRF composition. However, the integrity of the outer retinal layers and the dynamic changes of HRF in different AL stages may argue against degenerative changes (Piri et al., 2015). As leukocyte (especially macrophages) recruitment and activation occur in leukaemic niches (Ries et al., 2014), we speculate choroidal HRF may be leukaemic or inflammatory cells.
We did not find any difference in choriocapillaris FD density between AL and healthy eyes; similarly, the corrected FD density did not vary from the active to the remission phase of AL. These OCTA-based findings are consistent with the pathologic description from Allen & Straatsma (1961), who reported that choriocapillaris was uninvolved in patients with AL. We found an inverse correlation between choriocapillaris perfusion and the number of white blood cells. Assumptions on the relationship between choriocapillaris blood flow and systemic parameters are not straightforward, given the extensive interconnectivity of the choriocapillaris and the complex distribution of driving pressures and resistances throughout its network (Takita et al., 2018). However, we cannot exclude that the choriocapillaris perfusion was affected by hyperleukocytosis and blood flow disturbances due to the systemic condition (Cheung et al., 2021).
Only a few studies reported on the choroidal changes in blood dyscrasia. We also found an association between the choroidal thickness and the haemoglobin levels, in agreement with previous studies on patients with iron deficiency anaemia (Simsek et al., 2016) and β-thalassemia major (El-Shazly et al., 2016). We found an inverse association between the D-dimer and the choroidal vascularity in AL patients and a positive association between the D-dimer and the choroidal HRF count. Although a pro-coagulant state has been observed in patients with AL (Wang et al., 2018), it is difficult to hypothesize a comprehensive explanation combining choroidal vascular metrics (LCA and cFD density), the number of HRF and the D-dimer, and further research is warranted to confirm these putative associations.
This study has limitations, including the relatively small sample size and the lack of longitudinal analysis for all AL eyes. Our choroidal HRF quantification method required manual delineation of the choroidal area before imaging processing; automatic segmentation of the choroid and automated count of the HRF may improve the inter-rater reliability of the measurements. As we limited the analysis of the choriocapillaris FD density, the choroidal metrics and the choroidal HRF count to the central macula, widefield imaging could provide a more comprehensive picture of the degree of choroidal involvement in AL.
In conclusion, with the help of noninvasive multimodal imaging, we depicted in vivo the subclinical choroidal change in patients with AL, finding a predominant stromal involvement in the acute phase. We also identified choroidal HRF as a biomarker of leukaemic choroidopathy, and we hypothesized they could be leukocytes of either inflammatory or neoplastic origin. Choriocapillaris and choroidal vascularity correlated with a systemic procoagulant state, but further studies are needed to confirm and explain this relationship. Finally, a longer follow-up is warranted to elucidate the clinical value and the prognostic significance of leukaemic choroidopathy.

AU T HOR CON T R I BU T ION S
All the authors contributed to the conception or design of the work, the acquisition, analysis and interpretation of data, drafting the work, and revising it critically for important intellectual content. Each coauthor has seen and agrees with how his name is listed.