diff --git a/SIRF_data_preparation/Mediso_NEMA_IQ_lowcounts/README.md b/SIRF_data_preparation/Mediso_NEMA_IQ_lowcounts/README.md
new file mode 100644
index 0000000..b702c6f
--- /dev/null
+++ b/SIRF_data_preparation/Mediso_NEMA_IQ_lowcounts/README.md
@@ -0,0 +1,17 @@
+# Mediso_NEMA_IQ_lowcounts
+Aside from usual steps
+
+##
+Copy VOIs from high count data, but with adjustement of size. KT
+did this via STIR utilities (but can be done via SIRF as well)
+```
+for f in VOI*.hv; do
+    zoom_image --template data/Mediso_NEMA_IQ_lowcounts/OSEM_image.hv data/Mediso_NEMA_IQ_lowcounts/$f $f;
+done
+```
+## BSREM
+Given large oscillations with subsets, I reduced the number of subsets from 12 to 6 (in `dataset_settings.py`) but increased the initial step size.
+```
+python -m SIRF_data_preparation.get_penalisation_factor  --dataset=Mediso_NEMA_IQ_lowcounts --ref_dataset=Mediso_NEMA_IQ -w
+python -m SIRF_data_preparation.run_BSREM Mediso_NEMA_IQ_lowcounts --initial_step_size=.5 --relaxation_eta=.0105
+```
\ No newline at end of file