Implantable direct cardiac compression (DCC) systems such as our Heart Patch Pump can assist the failing heart without the risk of blood contact. To provide realtime, accurate support of pumping function, these devices need to assess hemodynamic variables such as cardiac output. For the Heart Patch Pump, we have based our control algorithm on time-varying stroke volume (SV(t)) estimates using sonomicrometer crystals placed in the heart. Multilinear regression was performed on 28 dimensions, during six different cardiac states. Then, six principal dimensions were identified as those with the greatest change during contraction. A strong correlation was maintained in all cardiac states when only the principal dimensions were used.